Introduction

Engaging in ADL is essential to everyone, as it allows them to maintain personal hygiene, physical health, and overall well-being1. Especially in children, ADL holds significant importance as they represent developmental milestones that must be achieved2. As part of their developmental journey, in early life, children need to learn about themselves by participating in personal tasks such as personal hygiene, dressing, and bathing. Then, they must engage in more complex tasks such as domestic activities like cooking and cleaning. This helps them understand their environment and family roles. As they continue to mature, they expand their interactions with the world by participating in community-related functions such as transportation, shopping, and managing money3. Participating in ADL not only fosters their holistic development2 but also develops essential learning skills4, such as information processing, motor learning5, attention, working memory, and problem-solving6,7. Moreover, practicing ADL also helps children strengthen their psychological abilities, such as self-reliance, self-confidence, and self-esteem. Those contribute to a positive life outlook and overall enhanced quality of life8,9.

Children with neurodevelopmental conditions such as autism spectrum disorder often struggle with performing everyday tasks due to the impact of their symptoms10. The level of difficulty varies depending on their individual spectrum11,12. The features of the autism spectrum that include challenges with social communication and interaction and repetitive behaviors serve as core diagnostic criteria. These difficulties often translate into struggles with asking for help and understanding instructions. However, these features are not only barriers to accomplishing daily routines. According to the Center for Disease Control and Prevention, a significant majority, ranging from 63% to 96% of individuals with ASD experience issues with sensory-motor integration and sensory sensitivity13,14,15. These challenges impact tasks requiring planning and executing motor actions and coordinating movements, such as buttoning clothes and tying shoelaces16,17. Some children with autism also struggle with executive function, impacting cognitive abilities like planning, organizing, and problem-solving. This can lead to a tendency to give up easily when facing challenges during ADL, as well as a lack of motivation to participate in the activities. A critical aspect of adapting to new situations or approaches is the ability to generalize18, a challenge often observed in individuals with autism19. For this reason, individuals on the autism spectrum require varying levels of support tailored to their severity and specific needs20.

Failure to achieve daily routines in individuals on the autism spectrum may be affected by various factors, including individual learning abilities, particularly challenges with generalization and limited access to learning resources and opportunities, such as a lack of personalized learning tools and insufficient technology integration. Developing individualized programs tailored to their needs can significantly enhance learning outcomes in ADL and support their overall development21 by providing learning opportunities, promoting independence, boosting self-esteem, and reducing stigma and discrimination22,23,24,25,26,27. The literature review suggested that designing effective personalized learning programs for children with autism should not only address their specific needs28 but also incorporate resources aligned with their interests and preferences. Strategies like utilizing teaching techniques, conducting task analysis to guide tasks, and consistently collecting data are recommended to uphold program integrity over time and optimize the child’s learning potential29. Although systematic analysis studies have indicated that traditional-based interventions, including educational and therapeutic approaches, hold significant potential to support children with autism, there are limitations across various domains, such as challenges in demographics due to the rising autism population and a shortage of therapists. Challenges in accessing in-person therapy sessions include transportation issues and costs. Traditional interventions often require a significant investment of time and resources from both the child and their caregivers. In the last decade, numerous technologies have been developed to address this challenge and provide support for children with autism and their families.

Technology has provided significant benefits for supporting individuals with autism across multiple dimensions. For example, several recent studies have successfully developed models for early diagnosis using various data forms, such as facial images and neuroimaging, by leveraging advanced machine learning and deep learning techniques12,30,31,32,33,34,35,36,37. While these advancements are crucial for diagnostic support, they represent a different domain of technological intervention. In contrast, our work focuses on a distinct, yet equally vital, challenge: the development of personalized learning programs for ADL.

Despite the growing body of research on technology for autism, existing ADL intervention often focus narrowly on specific ADL and may not cover all the skills necessary for their complete daily routines. For example, in teaching personal skills, such as showering, Kang et al. (2019)38 utilized the Kinect-based game “Take a Shower” to motivate children with autism to engage in showering tasks. The game was designed through task analysis of the showering process. After participating in the study for two weeks, six children showed an increase in the percentage of correctly performed steps in the task. To encourage children with autism to learn domestic skills, such as cooking. Merdan and Ozcan (2020)39 employed video modeling based on task analysis to teach them to cook pasta and fried eggs with garlic sausage. After participating in a one-on-one experiment for six days, three adolescents with autism showed improvement in their cooking skills. Additionally, Bereznak et al. (2012)40 used video prompting combined with task analysis to teach individuals with autism how to use a washing machine, operate a copy machine, and make noodles. After participating in the experiment, three teenagers with autism showed a significant improvement, achieving approximately 90% accuracy in correctly performed steps across all tasks, compared to their baseline performance, which averaged around 20%. Furthermore, in training children with autism in community-related skills, such as money management, Hassan et al. (2011)41 developed a digital game based on a storytelling concept. The game aimed to teach children with autism about various aspects of money management, including money identification, buying single and multiple items, and understanding money exchange. While nine children with autism who participated in the experiment showed improvement in money identification, they struggled with generalization. For example, most of them did not understand that the same amount of money could be used to purchase different types of items. The existing technologies are narrowly focused on a single task, which could make it challenging for them to transfer the learned skills to authentic environments. Moreover, current technological interventions often lack personalization and fail to account for the diverse needs and learning profiles of individuals with ASD. This lack of individualization can create barriers to engagement and successful skill acquisition.

From the review of existing works, it can be observed that most technologies have been designed with a primary focus on early diagnosis, aligning with the core principle that early intervention is critical for children. However, only a limited number of studies have focused on developing technologies to support training in ADL for children with autism, particularly those aimed at creating personalized learning programs that take into account the spectrum nature of the condition. Such programs should address specific needs, adapt to individual learning abilities, account for changes in baseline performance, and be sensitive to progress over time. The lack of this level of individualization can create barriers to both engagement and successful skill acquisition, such as generalization.

To address the limitations of existing interventions, we developed a Fuzzy logic-based tablet game application designed to enhance ADL skills in children with autism. The game was designed to promote their learning potential and accommodate the diverse needs of the autism spectrum. It featured six contexts reflecting daily life: Home, School, Transportation, Playground, Canteen, and Shopping. To ensure relevance and engagement, we employed the Participatory Design approach, collaborating with autism experts and users to co-create learning activities and storylines that align with the unique learning styles and needs of children with ASD. This approach ensured the design of gameplay, learning activities, and game mechanics aligned with the learning styles and unique characteristics of children with autism. Furthermore, we integrated a Fuzzy Logic-based Decision-Making System to provide personalized learning paths, dynamically adjusting the difficulty and content based on individual progress. This system supported decisions such as whether to increase the number of learning modules or maintain the same level for subsequent sessions. It also included rule-based algorithms to select appropriate learning modules to tailor learning experiences and optimize skill acquisition. The system’s storyline mirrors real-life scenarios, guiding children through daily routines from morning to bedtime. The purpose of this research was to evaluate the efficacy of this game in improving ADL skills in children with autism and skill generalization, which is often one of the most challenging aspects for children with autism. Additionally, the study examined the functionality of the Fuzzy logic-based decision-making system in designing personalized learning programs by tracking children’s learning progress while interacting with the game. A quasi-experimental, single-group pretest-posttest design was employed to examine the impact of a four-week intervention using the Fuzzy logic-based tablet game on enhancing activities of daily living skills in children with autism. The study also aimed to track learning progress using time-series data derived from game-based parameters. The experimental design was guided by the following hypotheses:

Hypothesis 1: Children with autism spectrum disorder who participate in the four-week intervention using the fuzzy logic-based tablet game will improve their ADL skills compared to their baseline performance, indicating enhanced skill acquisition and potential for generalization. This hypothesis reflected the game’s potential to improve skill generalization. To evaluate this, we compared pre- and post-intervention ADL scores reported by both parents and teachers using the non-parametric Wilcoxon Signed-Rank Test. These analyses were performed for overall ADL scores and for each of the six specific learning contexts to identify context-specific improvements in ADL skills, thus providing insights into potential skill generalization.

Hypothesis 2: The fuzzy logic-based decision-making system will effectively personalize the learning experience for children with autism by dynamically adjusting the number and difficulty of learning activities based on their individual performance, leading to distinct learning trajectories among different performance groups. To address this hypothesis, we analyzed the game-based data collected throughout the intervention to assess the functionality of the fuzzy logic-based decision-making system. We categorized participants into distinct learning performance groups based on the total number of learning modules completed within the first seven days and then examined the learning progress trends of each group over the four weeks. Learning trajectories (changes in the number of learning modules) between groups were compared using non-parametric tests, specifically the Friedman Test. In addition, the Independent-Samples Kruskal–Wallis Test, followed by post-hoc pairwise comparisons, was used to compare game parameters such as completion time and percentage accuracy, which served as indicators of learning performance and progress within each group.

Furthermore, we explored the relationship between the in-game reward system represented by the number of stars earned after completing each learning module and learning behaviors observed by teachers. The relationship was analyzed using Spearman’s rank correlation to suggest that the game’s assessment system provides a valid measure of learning engagement. Finally, the findings from all these statistical analyses were synthesized to provide a comprehensive evaluation of the game’s design and implementation as an innovative and engaging tool for teaching ADL skills and enhancing the learning potential of children with autism spectrum disorder, thereby addressing the broader aims of the study. Ultimately, this research contributes a novel educational tool that not only addresses the unique learning needs of children with autism but also offers practical support for parents and teachers. The game’s design and implementation present a viable framework for delivering effective, technology-based ADL interventions.

Results

This section presents the findings from the four-week intervention study evaluating the impact of the fuzzy logic-based tablet game on children with autism spectrum disorder. First, we describe the demographic characteristics of the participating children. Subsequently, we report the changes observed in their ADL skills as measured by pre- and post-intervention assessments completed by both parents and teachers. Then, we present the results pertaining to the functionality of the fuzzy logic-based decision-making system, including the categorization of participants into learning performance groups and the analysis of their learning trajectories based on in-game data. Finally, we report on the relationship between in-game performance and teachers’ observations of the participants’ learning behaviors during the intervention.

Participant demographics

The study was conducted across six schools: three special education centers and three mainstream schools (two with inclusive education classrooms and one with a mainstream integration classroom). One school was located in a metropolitan area, while the remaining five schools were in provincial areas. Initially, 45 children with autism participated (30 from special education centers, 10 from inclusive education classrooms, and 5 from the mainstream integration classroom). However, 29 participants were excluded due to incomplete participation in the four-week intervention or failure to meet the required session frequency. The final sample comprised 16 participants with a mean age of 10 years (SD = 2.1), ranging from 6 to 13 years.

Given the expected heterogeneity in learning abilities among individuals with autism, the fuzzy system was designed to personalize the learning program. This was particularly relevant in this study due to the wide age range of participants, highlighting the system’s potential adaptability across different developmental levels. To understand the participants’ baseline characteristics and learning potential, we analyzed data from paper-based assessments (Pre-ATEC, Gain-ATEC, and Gain-ADL) and game-based data, specifically the initial learning rate (as detailed in Table 1).

The pre-ATEC assessment results categorized participants’ autism symptom severity as mild (25%), moderate (50%), or severe (25%). These baseline characteristics indicated that while most participants exhibited mild levels of core diagnostic symptoms in the sociability and language/communication domains, many faced challenges in adapting to the new learning environment, particularly in foundational skills such as attention, understanding complex concepts, and self-motivation. This was supported by the sensory/cognitive awareness domain, where 43.75% of participants were classified as having moderate symptom severity.

To assess learning potential, we examined Gain-ATEC and Gain-ADL scores, as well as the initial learning rate. For Gain-ATEC, 62.5% of participants maintained their initial symptom severity levels. In terms of ADL skills (Gain-ADL), we defined a 10% improvement threshold based on a full score of 174, meaning participants had to achieve a gain of at least 17.4 points to be categorized as showing positive improvement. Six participants showed improvement, while ten maintained their abilities. The initial learning rate, calculated as the slope of the linear trend of the number of learning modules completed during the first seven days, reflects individual learning speed and adaptability to the game environment. The initial learning rate was used to classify participants into three learning groups: fast (31.25%), moderate (31.25%), and slow (37.5%). The analysis provided evidence of the wide range of learning performances within the study sample.

Table 1 The characteristics of the participants.
Full size table

Changes in activity of daily living skills

This research primarily aimed to evaluate the efficacy of the fuzzy logic-based tablet game in enhancing ADL skills in children with autism spectrum disorder, particularly addressing the challenge of skill generalization to different situations. To assess the game’s impact, we compared participants’ ADL skills before and after the four-week intervention across six contexts mirroring the game’s learning scenarios. However, the transportation context was excluded from this analysis due to participants’ lack of prior experience, which would have rendered the assessment unreliable. Overall improvement in ADL skills across all included contexts was analyzed using the non-parametric Wilcoxon Signed-Rank Test. As shown in Table 2, a statistically significant improvement was observed in the total ADL score (\(p < 0.05\)). At the individual context level, a significant improvement was found specifically in the canteen context (\(p < 0.01\)). While other contexts did not reach statistical significance at the conventional threshold, a trend of improvement was evident across all contexts when analyzing the percentage of change relative to each participant’s individual potential for improvement. This potential was defined as the difference between their baseline performance and the maximum possible score in each context. The observed progress, calculated based on this individual potential, was most notable in the shopping context (20.15%), followed by the home context (15.10%), school context (5.95%), and playground context (5.47%).

These findings suggest that the fuzzy logic-based tablet game holds promise as an effective tool for promoting ADL skills in children with autism spectrum disorder. The observed improvements, even when not statistically significant in all individual contexts, indicate the game’s potential to address difficulties often faced by this population in generalizing learned skills across different situations, thereby supporting the expansion of their functional abilities.

Table 2 Changes in Activities of Daily Living (ADL) Scores (Wilcoxon Signed-Rank Test) and Percentage Improvement Relative to Individual Potential, by Context.
Full size table

Functionality of the fuzzy logic-based decision-making system and learning trajectories

The learning profiles of individuals with autism spectrum disorder are highly heterogeneous, influenced by variations in cognitive and processing skills (e.g., attention span, working memory, detail-focused processing, and generalization), psychological states (e.g., interest, motivation, and engagement), prior experiences (e.g., domain-specific knowledge and learning history), and access to individualized enriched learning environments. To address this inherent variability, we developed a fuzzy logic-based tablet game incorporating a Fuzzy Expert System to create a personalized learning program. Given the established impact of the autism spectrum’s diverse characteristics on learning outcomes, encompassing both cognitive and psychological factors, this section aims to elucidate the characteristics of participants who demonstrated effective learning through our intervention, building upon our previous findings of improved ADL skill generalization. Furthermore, we sought to investigate how personalized learning, facilitated by the Fuzzy Expert System, might influence different learners’ learning trajectories and adaptive skill development. To achieve these aims, we categorized participants into distinct learning groups based on their initial learning rate within the game. Subsequently, we compare the learning outcomes achieved by these different groups. Finally, we track and analyze the game-based learning trajectories of these groups over the course of the intervention, utilizing key in-game performance parameters to understand their progression.

Participant categorization by initial learning rate

To categorize participants into groups, we initially explored using pre-ATEC scores, which assess autism symptom severity. While overall pre-ATEC scores indicated that participants have moderate severity with relatively mild social and communication challenges, they also revealed great difficulties in cognitive skills crucial for learning. This suggested that pre-ATEC alone might not be a reliable indicator of learning potential, as it does not fully capture psychological factors like interest and motivation. We also considered generalization of ADL skills using gain-ADL scores, categorizing participants into positive improvement or no change. However, this measure provided a post-intervention snapshot rather than insight into learning progress over time.

To address these limitations and to directly reflect participants’ initial engagement and learning within the game environment, we ultimately classified them based on their Initial Learning Rate. This rate was determined by analyzing the trend of learning module completion during the first seven days, reflecting how quickly participants adapted to and interacted with the game. This approach also indirectly captured the initial effectiveness of the Fuzzy Expert System in providing personalized learning paths, as all participants started with the same number of modules, and subsequent progression was based on their performance. We found the Initial Learning Rate to be a more comprehensive metric for grouping learners, as it integrated aspects of cognitive skills, psychological engagement within the game context, and early learning experience in the structured intervention.

We attempted to classify participants into learning performance groups by first considering their pre-ATEC scores, which assess participants’ characteristics in terms of autism symptoms. Since symptom severity can impact individual learning abilities, we used these scores as an initial reference. The overall pre-ATEC scores categorized participants as having moderate autism severity, they exhibited mild symptom severity in core domains such as social and communication skills. However, they faced greater challenges in cognitive skills, which are fundamental to learning. This finding suggests that using the overall ATEC score may not accurately reflect a child’s learning abilities, as it does not account for psychological factors such as interest and motivation, which play a crucial role in learning. To further refine classification, we considered generalization skills related to ADLs, measured by the overall gain-ADL score. We defined a 10% improvement threshold based on a full score of 174, meaning participants had to achieve a gain of at least 17.4 points to be categorized as showing positive improvement, no change, or negative improvement. However, the gain-ADL score could only classify participants into two groups: positive improvement and no change. While this measure indicated how well participants performed after engaging with the game, it did not provide insight into their progress over time. To address this limitation, we incorporated game-based data, particularly the number of learning modules completed over time during the first seven days. We analyzed the trendline values to assess how quickly participants learned and interacted with the game. These trendline values also reflected the effectiveness of the Fuzzy Expert System in adapting and designing personalized learning paths for children with autism. In our personalized learning path design, all participants started with four learning modules, and the number of modules changed based on their performance. We found this approach to be more effective than using only Pre-ATEC or gain-ADL scores, as it directly reflected learning foundations, including cognitive skills, psychological factors, and learning experience within the same structured environment.

To categorize participants based on their initial learning rate, defined as the slope (’a’ value) of the linear trend in learning module completion over the first seven days (\(y = ax + b\)), we established three groups: fast, moderate, and slow learners. The classification criteria were as follows: the fast learning group (FLG) comprised participants with an ’a’ value at least 0.5 standard deviations (SD) above the total mean, indicating rapid adaptation to the game. The slow learning group (SLG) included those with an ’a’ value at least 0.5 SD below the total mean, signifying gradual adaptation. Participants with ’a’ values falling between these thresholds were classified into the moderate learning group (MLG). Figure 1 illustrates the increasing number of learning modules completed by participants in each of these groups over the first seven days, visually representing their initial learning trajectories. As shown in Fig. 1, the FLG (green lines) and MLG (orange lines) each consisted of 5 participants, with average ’a’ values of 0.85 and 0.42, respectively. The SLG (red lines) included 6 participants, with an average ’a’ value of 0.07. Statistical analysis of these ’a’ values using the Friedman Test revealed a significant difference between the groups (\(p =.007\), \(p < 0.01\)), with within-group consistency also being significant (\(p < 0.001\)). These initial learning rates demonstrated a clear distinction in learning speed during the first week, with the fast learning group progressing more quickly than the moderate and slow learning groups. However, it is important to note that all participants showed positive learning trends over time, with the slow learning group demonstrating substantial progress and eventually achieving a similar total number of completed modules as the other groups.

Fig. 1
figure 1

Initial learning trajectories: increase in the number of learning modules completed during the first week of intervention for fast learning group (FLG - green lines), moderate learning group (MLG - orange lines), and slow learning group (SLG - red lines).

Full size image

As illustrated in Fig. 2, participants within each of the three learning rate groups (FLG, MLG, and SLG) exhibited a mix of pre-intervention autism symptom severity (based on pre-ATEC scores) and post-intervention changes in both autism symptoms (Gain-ATEC) and ADL skills (Gain-ADL). This heterogeneity across learning speeds suggests that the Fuzzy Expert System was effective in adapting to a diverse range of learning profiles and behaviors, rather than solely catering to a specific symptom severity level. Furthermore, this distribution provides evidence that technology-based interventions can support learning and promote performance gains in children with autism across the spectrum.

Fig. 2
figure 2

Heterogeneity in pre-intervention autism severity and post-intervention gains in ATEC and ADL across learning rate groups.

Full size image

Examining each group in more detail: The Fast Learning Group (FLG) generally showed higher learning engagement consistent with their paper-based assessments both before and after the intervention. A majority (60%) presented with mild symptom severity at baseline, and all maintained this level post-intervention. Notably, 60%of the FLG also demonstrated positive improvement in ADL skills, with no participants showing negative changes. The Moderate Learning Group (MLG) displayed a more varied pattern, aligning their game-based learning with their human-scored assessments. At baseline, 60% exhibited moderate symptom severity. Post-intervention, their outcomes were mixed: two showed no change in severity, two improved, and one showed a negative change. Regarding ADL skills, 60% maintained their baseline abilities, while one participant showed positive improvement. Despite the Slow Learning Group (SLG) demonstrating a slower initial learning rate in the game, they also showed evidence of learning. At baseline, 66.66% had moderate symptom severity. Post-intervention, half showed no change in severity, and two participants demonstrated positive improvement, with only one participant showing negative change. This finding suggests that a slower initial learning rate, as captured by game data, does not necessarily equate to a low capacity to learn through technology-based interventions. It highlights the need for diverse learning tools and approaches to support the varied learning styles and paces of children with autism.

Learning performance comparisons among fast, moderate, and slow learning groups

To assess the distinct learning characteristics of the fast, moderate, and slow learning groups, we compared their game-based performance data, specifically overall percentage accuracy and completion time. Non-parametric Independent-Samples Kruskal-Wallis Tests were conducted across the three groups, followed by pairwise comparisons to identify significant differences.

As presented Figure 3 (Right), a significant main effect was observed for percentage accuracy across the groups (\(p < 0.001\)). Post-hoc pairwise comparisons further confirmed significant differences between all three groups (\(p < 0.001\) for all pairs). The mean percentage accuracy (and standard deviation) for each group was: FLG = 89.6% (SD = 13.9), MLG = 79.7% (SD= 21), and SLG = 60.5% (SD = 25.8). Similarly, Fig. 3 (Left) illustrates a significant overall difference in completion time among the groups (\(p < 0.001\)). Pairwise comparison tests also revealed significant differences across all three group pairings (\(p < 0.001\) for all pairs). The mean completion time (and standard deviation) for each group was FLG = 1.70 sec (SD = 1.54), MLG = 3.43 sec (SD = 3.42), and SLG = 5.50 sec (SD = 4.89). These findings collectively confirm that the three learning groups, categorized by their initial learning rate, exhibited statistically distinct game-based learning performance. The observed differentiation in both accuracy and completion time suggests the Fuzzy Expert System’s effectiveness in accurately responding to children’s individual performance levels and adapting the learning environment, potentially indicating the provision of a personalized and appropriately challenging learning experience.

Fig. 3
figure 3

Comparative learning performance: box and violin plots of (Left) percentage accuracy and (Right) completion time displaying the distinct outcomes for fast learning group (FLG), moderate learning group (MLG), and slow learning group (SLG).

Full size image

Game-based learning trajectories of learning groups

Building upon the demonstrated ability of the Fuzzy Expert System to create personalized learning programs based on individual performance, we further examined the learning trajectories of the three categorized groups (FLG, MLG, SLG) over time. This analysis aimed to understand how participants’ learning progressed within the adaptive environment, particularly in terms of their percentage accuracy and completion time. The inherent mixed performance within each group, influenced by cognitive skills and psychological abilities, makes this longitudinal tracking crucial for assessing the sustained effectiveness of technology-based interventions.

Figure 4 illustrates the learning trajectories of each group, incorporating two y-axes to provide a comprehensive view: percentage accuracy (left axis) and the number of learning modules (right axis). The shaded areas also indicate increasing task difficulty, directly correlating with the growing number of learning modules assigned by the Fuzzy Expert System. For percentage accuracy, boxplots display the daily performance range, with the median value denoted by a line within each box. Average daily values are plotted as distinct points over the boxplots for comparison. A linear trendline (\(y = ax + b\)), based on the median and depicted by a black dashed line, shows the overall learning trend. In this linear equation, the ’b’ coefficient value represents the estimated performance at the beginning of the observation period. The ’a’ coefficient value indicates the rate of change over time (slope), with a positive ’a’ signifying improvement in accuracy (i.e., increasing percentage).

Crucially, the trendlines reveal positive improvement trends across all learning groups. The Fast Learning Group (FLG) began with the highest initial percentage accuracy (\(b=92.18\)%), followed by the Moderate Learning Group (MLG) at 78.06%, and the Slow Learning Group (SLG) at 53.30%. Despite their lower starting point, the SLG demonstrated the steepest improvement rate (\(a=0.792\)), significantly surpassing the MLG (\(a=0.448\)) and FLG (\(a=0.236\)). This indicates that while FLG participants maintained strong performance even with increasing task difficulty, SLG participants showed remarkable capacity for growth and significant catch-up within the personalized learning environment.

For completion time, the data are visualized using the same graph format as percentage accuracy. However, the shaded areas in this plot represent the increasing challenge participants encountered daily, determined by two main components: the number of learning modules completed over time and the Task Complexity Index (TCI). The TCI quantifies the difficulty level of each learning module by incorporating three weighted factors relevant to autistic learners: Cognitive Load (CL, 45%), measured by the number of steps per module, Interaction Style (IS, 35%), reflecting the complexity of user engagement, and Task Context (TC, 20%), based on the familiarity of the scenario (self-care, domestic, or community-related tasks). The Overall Daily Complexity was computed with equal weight between the number of learning modules earned and the task complexity index. This metric allowed us to quantify and track the adaptive learning demands each group experienced, reflecting the Fuzzy Expert System’s ability to tailor challenges to individual performance.

The overall results for completion time indicated a positive trend of improvement across all learning groups, evidenced by the downward slope of their trendlines. The FLG demonstrated the best initial performance, with the lowest intercept (\(b=1.53\) seconds), followed by the MLG at 2.71 seconds, and the SLG at 4.73 seconds. Despite starting with longer completion times, the SLG exhibited the greatest rate of improvement, reflected in the steepest negative slope (\(a=\text {-}0.138\)), surpassing the MLG (\(\text {-}0.050\)) and FLG (\(\text {-}0.032\)). This suggests that the adaptive learning environment, driven by the Fuzzy Expert System, effectively supported each group’s unique learning trajectory. For children with autism, who often benefit from individualized pacing, predictable structure, and tasks matched to their processing capabilities, this adaptive approach appears crucial in fostering engagement and incremental success, even for those with lower baseline abilities.

These longitudinal analyses provide evidence of the Fuzzy Expert System’s capability to promote and support learning across diverse initial abilities, highlighting the continued progress of all groups within the personalized adaptive framework.

Fig. 4
figure 4

Game-based learning trajectories of learning groups: (Left) percentage accuracy progression for fast learning group (FLG), moderate learning group (MLG), and slow learning group (SLG), respectively, over the intervention period. (Right) completion time progression for FLG, MLG, and SLG, respectively, over the intervention period. Each plot displays individual median data lines and the linear trendline fitted to the group mean data.

Full size image

Relationship between in-game performance and observed learning behaviors

To validate the game’s internal performance metrics against external expert observations, the in-game star rating system, which was designed to reflect the child user’s performance, was correlated with children’s real-world learning behaviors as perceived by teachers. Teachers evaluated foundational learning skills, including attention, motivation, and engagement, using the Learning Behavior Observation Form. Spearman’s rank correlation analysis was performed between the star ratings earned in the game and the teachers’ evaluations. The results indicated strong positive correlations. For the overall participant group, a correlation of \(r=0.62\) was observed. When examining individual learning groups, the Fast Learning Group (FLG) showed a correlation of \(r=0.75\), while the Slow Learning Group (SLG) exhibited a correlation of \(r=0.80\). The Moderate Learning Group (MLG) demonstrated a correlation of \(r=0.44\). These findings provide compelling evidence that the game’s star rating system effectively serves as a valid and reliable indicator for measuring children’s learning behaviors during the intervention, bridging game-based performance with real-world expert observation.

Discussion

This study aimed to evaluate the efficacy of a fuzzy logic-based tablet game in enhancing Activities of Daily Living (ADL) skills and promoting generalization among children with autism spectrum disorder. A parallel objective was to assess the functionality of the game’s integrated fuzzy logic-based decision-making system in providing personalized learning experiences. To address these aims, the research formulated two primary hypotheses: (1) that the game intervention would lead to improved ADL skills and enhanced generalization in participants; and (2) that the fuzzy logic-based decision-making system would effectively personalize learning, resulting in distinct learning trajectories and that its internal metrics would correlate with observed learning behaviors. This section will now discuss the study’s findings in relation to these hypotheses and the broader implications of the intervention. First, we will analyze the enhancement of activities of daily living skills and generalization, delving into the observed improvements in functional abilities. Subsequently, the discussion will focus on the fuzzy logic-based decision-making system: personalization and learning trajectories, interpreting its role in adaptive learning, and the observed learning progress. Finally, we will explore addressing learner diversity through personalized intervention, examining how the game catered to the varied characteristics of the participants and the implications for broader applicability.

Enhancement of activities of daily living skills and generalization

The primary objective of this study, as articulated in Hypothesis 1, was to enhance participants’ ability to perform ADLs and promote skill generalization among children with autism spectrum disorder. This objective directly addresses a significant challenge for individuals with autism: applying learned skills across various contexts. Our findings provide empirical evidence supporting the effectiveness of the fuzzy logic-based tablet game in enhancing daily routine skills.

Post-intervention ADL assessments revealed a statistically significant improvement in participants’ overall ADL skills (\(p<0.05\)), with a particularly notable gain observed in the canteen context (\(p<0.01\)) (Table 2). While other contexts (shopping, home, school, playground) did not reach conventional statistical significance, they consistently demonstrated positive trends in improvement, reflecting individual progress against their potential for skill acquisition. This suggests the game’s capacity to foster incremental gains across a range of daily tasks.

These results align with and expand upon previous research demonstrating the utility of technology-based interventions for ADL skill development in autism. For instance, while studies like Kang et al.38(showering), Merdan and Ozcan39 (cooking), and Bereznak et al.40 (washing machine use) successfully taught specific personal or domestic skills, they often focused on single tasks. Our multi-context game, covering various aspects of a child’s day (home, school, canteen, playground, shopping), extends this work by promoting a broader range of ADLs. Similarly, interventions for community-related skills, such as money management414243, have shown promise but often noted challenges with generalization beyond the specific trained scenario. The observed positive trends across multiple diverse contexts in our study suggest that the game’s holistic approach to simulating daily routines may contribute more effectively to the generalization of ADL skills in varied real-life situations. The integration of engaging elements, mirroring findings by Merdan and Ozcan39 regarding interest and motivation, likely played a crucial role in fostering sustained participation, which is vital for generalization.

The observed improvements in ADL performance lead to the conclusion that this technology-based intervention serves as a viable tool for promoting skill generalization, a critical aspect for functional independence, social interaction, and overall quality of life in individuals with autism. The progress observed across all three learning groups (Fast, Moderate, Slow), as indicated by ADL gain scores (Fig. 2), further reinforces that participants, regardless of their baseline symptom severity or initial learning rate, were capable of making progress within this technology-based intervention. This highlights the importance of engagement and personalized adaptation in facilitating the understanding of abstract concepts, generalization of rules, and application of problem-solving strategies across different tasks and contexts.

The game’s effectiveness in promoting ADL skills, particularly in supporting generalization, validates the comprehensive design and development approach employed. The Participatory Design (PD) methodology was instrumental in this success, ensuring that the game’s rich learning environment, realistic contexts (Home, School, Canteen, Transportation, Playground, and Shopping), and 38 learning activities were directly relevant and applicable to everyday situations. This approach, where autism experts and target users actively co-created the intervention, is consistent with systematic reviews highlighting PD’s ability to enhance engagement, skill development, and individualized support in technology for children with autism44. By integrating expert knowledge of autism learning styles with user feedback on engagement and realism, the PD process (Sturm et al.45, Malinverni et al.46) directly contributed to designing a game that effectively facilitated the transfer of skills to real-world scenarios, a key challenge in generalization.

The primary objective of the game was to enhance participants’ ability to perform ADLs. We assessed ADL performance by comparing pre- and post-intervention scores. This evaluation also reflects the participants’ ability to generalize skills, which is a significant challenge for children with autism when applying learned skills from one context to another. The ADL score results demonstrated that the study successfully achieved its purpose and added empirical evidence to support the effectiveness of technology-based interventions in enhancing learning for children with autism, particularly in performing daily routines. Participants showed significant improvement in ADL skills across all contexts, with statistical significance at \(p < 0.05\). Additionally, Table 2 indicated significant improvement, specifically in the canteen context, with statistical significance at \(p < 0.01\). Although the other contexts did not show statistically significant improvements, there was a noticeable trend of improvement: 20.15% in the shopping context, 15.10% at home, 5.95% at school, and 5.47% at the playground. This aligns with existing technologies designed for ADLs skills in autism. For personal tasks such as showering, Kang et al.38 successfully taught children with autism the steps using a Kinect-based game over 21 sessions within 11 weeks. This intervention included visual modeling to help participants practice showering while interacting with the game. In domestic tasks like cooking, Merdan et al.39 observed positive improvements in daily life skills such as cooking among children with autism after participating in a study that used video modeling and task analysis to teach cooking techniques. The researchers noted that these methods could foster interest, motivation, self-assessment abilities, and self-confidence in children with autism, potentially increasing their participation in ADL. For community-related functions such as understanding money concepts and purchasing skills, various techniques have been employed. These include tablet games (Buyuk et al.42), motion-controlled games (Chiu et al.43), and digital storytelling applications (Hassan et al.41). The content covered money identification and learning money exchange through a digital storyline that progressed from simple tasks to more advanced ones. The storytelling approach provided children with visual materials, aiding those with autism in learning effectively. The results demonstrated that children with autism could apply the knowledge gained from the game to real-life situations such as making actual purchases.

The improvement observed in participants’ performance in ADL led us to conclude that technology-based interventions, such as this game, are as effective as traditional interventions in promoting skill generalization. Generalizing skills is a common challenge for many children with autism. It is crucial for several reasons including promoting functional independence, social interaction, learning and adaptation, and behavioral flexibility. The ability to generalize skills is associated with better long-term outcomes such as improved academic achievement, vocational success, and overall quality of life. Evidence from Table 2 showed that the game enhanced participants’ ability to function independently in various real-life situation such as home, school, and community. Additionally, positive improvements were observed among participants across all three groups, particularly when considering the 10% change, as shown in Fig. 2. This suggests that regardless of their severity level or initial learning ability, participants were capable of learning and making progress, especially when engaged in the technology-based intervention. Psychological factors such as interest, motivation, and engagement appeared to play a significant role in enhancing their learning abilities. Furthermore, the results indicate that participants were able to understand abstract concepts, generalize rules, and apply problem-solving strategies across different tasks and contexts.

The game’s effectiveness in promoting ADL skills, particularly in supporting generalization, validates the design and development approach we employed. This included the Participatory Design approach, in which experts and volunteer participants from the target user group were actively involved in the design process. The results highlighted the value of the rich learning environment we created, featuring contexts and learning activities grounded in real-life scenarios. This allowed children with autism to learn and practice their skills in ways that are directly applicable to everyday situations. We designed 6 learning contexts with a total of 38 learning activities. The design of these contexts was based on the hypothesis that children should learn to prepare themselves for real-world situations they might encounter. We used Participatory Design as a tool for fostering collaboration between experts and user volunteers. This approach was supported by strong evidence from a systematic literature review by Maun et al.44, which indicated that Participatory Design could help children with autism by increasing engagement, developing skills, and supporting individual needs. In this work, we employed techniques that involved experts and user volunteers working together in the design process. Experts were responsible for designing the game story, creating the game storyboard, approving the graphic design, developing the game’s storyline, and defining game elements such as roles and rewards. User volunteers supported the experts in designing the game’s storyline. This approach aligned with the successful application of PD by Malinverni et al.46, where experts were involved in designing the game’s mechanism structure, such as player interaction patterns and the implementation of therapeutic goals. Additionally, children with autism contributed to designing game mechanism specifications, such as rewards and feedback. This collaborative effort resulted in positive outcomes including high rates of acceptance and engagement from children with autism. We also invited user volunteers to play the demo version of the game under the supervision of their therapist. We gathered information to improve the game by observing the children’s interactions and interviewing the therapists after the sessions. This approach was similar to the work of Sturm et al45 who invited children with autism to play the game and provide feedback to enhance the game’s motivation and engagement capacity.

The fuzzy logic-based decision-making system: personalization and learning trajectories

This section addresses Hypothesis 2, which states that the fuzzy expert system would effectively design personalized learning programs for children with autism and lead to distinct, yet positive, learning trajectories. The fuzzy expert system served as a dynamic decision-making tool, adjusting the number of learning modules and selecting appropriate activities based on each child’s real-time performance within the game, thereby creating a truly individualized learning experience. As illustrated in Fig. 1, the system successfully adapted to the diverse learning profiles of the participants, demonstrating its capacity to respond effectively to individual needs and behaviors.

Our analysis of game-based data, specifically the number of completed learning modules, led to the classification of participants into three distinct groups: fast, moderate, and slow learners. It is important to note that while paper-based assessments provided valuable baseline information, they alone could not fully capture the nuances of individual characteristics, interaction patterns, or holistic learning progress. This finding aligns with Willis et al. (2021)47, who highlighted how technology-based assessments can complement traditional methods to provide a more comprehensive understanding of individuals with autism.

To validate the accuracy of the fuzzy expert system in assessing individual performance, we analyzed key in-game parameters, including percentage accuracy and completion time, using the non-parametric Independent-Samples Kruskal-Wallis Test. We found significant differences in performance among the three learning groups (\(p<0.001\)), as detailed in Fig. 3. The Fast Learning Group (FLG) consistently demonstrated superior task memory for each daily routine task, quicker responses, and higher accuracy, suggesting more efficient cognitive processing. However, a crucial finding was that all three groups ultimately demonstrated the capacity to reach their learning potential, as evidenced by the time-series data in Fig. 4, which illustrates their positive learning trajectories. For instance, even though the FLG started with a high percentage accuracy, they still improved their performance, achieving a rate of change over time of 0.446. On the other hand, the slow learning group started with a low percentage accuracy but showed a remarkably higher rate of change at 1.17. Similarly, the FLG outperformed others in initial completion time and rate of decrease. These results strongly confirm the fuzzy expert system’s capability to accurately assess performance and tailor personalized learning programs for each child, underscoring that not only cognitive abilities but also factors like interest, motivation, and engagement significantly influence learning outcomes in children with autism.

Further emphasizing the holistic approach, we incorporated insights from supervising teachers. Their questionnaire ratings on children’s attention, motivation, and engagement provided valuable qualitative data on task initiation, sustained attention, and enjoyment. We integrated a star reward system within the game, designed to reflect these same teacher-assessed areas. We found a strong correlation between teacher ratings and the game-based star system (overall participants: \(r=0.62\); FLG: \(r=0.75\); Moderate Learning Group [MLG]: \(r=0.44\); SLG: \(r=0.80\)). This correlation reinforces the notion that levels of interest can significantly impact learning behavior (Koegel et al.,)48. Similarly, Yeung and Chan49 found that impairments in executive function are related to motivation and emotional function. Difficulties in these areas can impact an individual’s ability to initiate and sustain activities, manage frustration, and shift attention from preferred activities to less preferred ones. Despite these variations, the observed positive learning trajectories across all groups affirm the system’s effectiveness in guiding learners toward their objectives. This approach also nurtures a positive mindset and fosters self-determination along the way.

Based on the discussion above, the Fuzzy Expert System, leveraging fuzzy logic, offers adaptability, holistic understanding, and flexibility in assessing each child’s performance and creating personalized learning environments that harness individual potential. Despite the varying severity levels observed in children with autism, this adaptable and interest-aligned framework proved effective in fostering learning outcomes across varying symptom severities. The versatility of fuzzy logic positions it as a valuable tool for researchers and educators committed to supporting children with autism, consistent with its documented success in assessing engagement50 and communication skills51. Our comprehensive data collection, combining paper-based and game-based assessments, offered a detailed understanding of participants’ learning behaviors and the effectiveness of the system’s progressive increase in learning activities compared to alternative data collection methods.

Conclusion and limitation

The study illuminated the diverse abilities exhibited by children across the autism spectrum, encompassing a range of symptoms. While standardized assessments were adept at categorizing individuals based on the severity of their symptoms, they may fall short in capturing nuanced psychological traits such as interests, motivation, and engagement, particularly in the realm of technology-based interventions. Our research focused on developing and evaluating a tablet-based game specifically designed to promote ADL skills and enhance learning for children with autism, integrated with a tailored fuzzy expert system. The results showed that the tablet-based game has the potential to improve participants’ learning abilities and ADL skills, particularly when implemented under the supervision of specialists such as teachers. We successfully established an effective methodology for categorizing participant performance, with game-based data proving to be the most accurate indicator of individual capabilities and consistency. This allowed us to categorize participants based on their adaptability within the game environment, reinforcing the game’s alignment with personalized, inclusive learning interventions that promote independence. This research contributed valuable insights into the development and evaluation of effective game applications for enhancing ADL skills in children with autism. However, it was crucial to acknowledge certain limitations. As this research was the pilot study, it was primarily designed to examine the feasibility and preliminary efficacy of the fuzzy logic-based tablet game in promoting ADL skills among children with autism. While the findings were encouraging, several limitations should be acknowledged. These include the small number of participants, the short duration of the intervention, and the heterogeneity of participant characteristics, performance levels, and ages. The study setting also imposed restrictions: because it was conducted in a rural area with limited transportation access, the transportation-related learning context had to be excluded from the analysis. In addition, the use of a quasi-experimental, single-group pre–post design further limited the strength of the experimental conclusions.

Methods

Fuzzy logic-based tablet game application

We developed “The Adventures of Imboon and Oonjai,” the Fuzzy logic-based tablet game in Thai language designed to enhance ADL skills in children with autism. In this game, children choose to play as either Imboon (male) or Oonjai (female) and guide their chosen character through daily routines, from waking up in the morning, brushing teeth, taking a shower, engaging in various activities, and finally going to sleep at the end of the day. By assisting the character with each step, children learn to perform the corresponding ADL tasks themselves through interactive activities such as dragging and selecting items. The game provides support through tutorials for each new task, ensuring children have the opportunity to learn and practice before attempting it independently. Developed with the principles of inclusive education in mind, the game emphasizes individualized learning and real-life scenarios to promote engagement, skill acquisition, and generalization.

To ensure our game effectively met the needs of children with autism, maximized their learning potential, and promoted generalization to real-life settings5253, we employed a Participatory Design (PD) approach45. This collaborative methodology, successfully utilized in previous technology designs for children with autism4445465455, actively involved key stakeholders throughout the game development process. The PD process in this study was central to the learning activity design and testing phases. We collaborated closely with a multidisciplinary team of experts in autism, including an occupational therapist, a developmental psychologist, a special needs educational teacher, and educational technology specialists, alongside the target users themselves (children with autism). This collaboration ensured the creation of engaging and relevant learning activities, each carefully broken down into sequential steps using task analysis approach. This process resulted in a game encompassing six real-world contexts – Home, School, Transportation, Playground, Canteen, and Shopping (Fig. 5) – featuring a total of 38 learning activities. In the testing phase, we invited three children with autism and their occupational therapists. The occupational therapist served as mentors and provided recommendations for improving the game based on their real-world experiences. They provided valuable feedback for enhancing the game’s flow, smoothness, and ability to maximize learning potential in children with autism. Specifically, they highlighted the importance of maintaining engagement during transitions between learning activities. Based on their recommendations, we refined the game by expanding interactive areas, incorporating animations during loading and transitions, and adding text and voiceovers for clearer guidance to guide children as they helped the character complete tasks.

Fig. 5
figure 5

The six learning contexts in the Fuzzy Logic-based Tablet Game (The Adventures of Imboon and Oonjai).

Full size image

To provide personalized learning paths, we integrated a fuzzy logic-based decision-making system into the game’s storyline, which mirrors real-life daily situations for children. This system aimed to enhance motivation and ultimately improve learning outcomes by tailoring the gameplay difficulty based on individual performance. The fuzzy logic-based decision-making system employed a hybrid approach, combining fuzzy logic and rule-based approach. Unlike traditional logic that deals with strict “true” or ‘false” conditions, fuzzy logic allows for “degrees” of truth. This is particularly useful for understanding how children with autism learn, as their skills often develop gradually and with variations that do not always fit into simple categories, reflecting the spectrum of their behaviors, which can be influenced by various factors, including psychological states. In this study, the fuzzy logic component is used to analyze a child’s performance and behavior within the game, evaluating factors like accuracy and completion time. Instead of making a binary judgment of “good” or “bad,” fuzzy logic allowed the system to assess the degree to which a child was performing well. Based on this nuanced evaluation, the fuzzy logic rules adaptively determine whether to maintain the current number and difficulty of learning activities or to increase them for the subsequent session, aiming to provide an appropriately challenging and engaging experience tailored to their individual learning trajectory. Complementing this, the rule-based approach was responsible for selecting specific learning activities from the curriculum to construct a cohesive and contextually relevant learning storyline that aligns with the children’s everyday lives and the overall learning objectives.

Building upon the concept of personalized learning facilitated by fuzzy logic, the architecture of our fuzzy logic-based decision-making system, illustrated Fig. 6, operates as follows. Children’s in-game performance was evaluated through a set of membership functions and fuzzy rules, which translated quantitative game data into qualitative learning categories. As the child interacted with the tablet-based game, performance in various learning activities was continuously fed into the fuzzy logic-based decision-making system. Within this system, the fuzzifier processed performance parameters (e.g., accuracy, completion time, and percentage of hints used) and transformed them into fuzzy input sets through predefined membership functions. Each variable was represented by linguistic categories; for example, accuracy and hint usage were classified into three categories (low, medium, and high) using triangular membership functions spanning the observed data range. Accuracy above 75% was assigned to the high membership, while accuracy below 25% was considered low. Next, fuzzy rules combined these input variables to determine the child’s learning status. For example: “IF accuracy is high AND hint usage is low, THEN learning ability is classified as good.” And “IF accuracy is low AND hint usage is high, THEN learning ability is classified as low.” These rules enabled the system to dynamically assign participants into learning groups based on the number of learning modules completed during gameplay. The output of the inference engine was a fuzzy output set, which was then defuzzified into a crisp decision. This decision specified whether to increase or maintain the number of learning modules for the next session, directly adjusting the challenge level and learning pace.

The decision from the fuzzy logic-based decision-making system then guided the learning module selection process. Based on the fuzzy output and predefined rules, the system determined which learning module within a given context (e.g., Home, School, Canteen) would be presented next, thereby adapting the environment and driving game progression. To ensure children effectively learned ADL steps, the system also decided how many days a child needed to repeat a module to consolidate skills, while avoiding excessive repetition that could cause boredom. Additionally, modules were scheduled for replay after a delay, functioning as a “retest” to evaluate retention of ADL steps. To support smooth transitions, video-based modules were integrated between active learning tasks. These non-interactive modules helped reduce cognitive load and prepared children for the next activity, as well as to ensure that the game’s daily storyline aligned with real-life scenarios. As a result, each child followed an individualized learning path shaped by their performance profile. Furthermore, throughout this adaptive process, the game incorporated various interactive elements–feedback, sound, rewards, challenges, narrative features, and interactive mechanics such as dragging and selecting–to sustain engagement and optimize learning outcomes.

Fig. 6
figure 6

The workflow diagram of the Fuzzy Logic-Based Decision-Making System.

Full size image

Evaluation of the fuzzy logic-based game intervention

To evaluate the efficacy of the fuzzy logic-based tablet game in enhancing ADL skills in children with autism spectrum disorder, a pilot study was conducted employing a quasi-experimental, single-group pre-post test design to examine the impact of a four-week intervention using the Fuzzy logic-based tablet game on promoting ADL skills. ADL improvements were assessed through human scoring, and learning abilities were tracked via in-game data collection. This design was chosen for its feasibility in evaluating the intervention’s effects within the children’s natural learning environments. The experiment was conducted in schools, where children spend most of their time, and data were collected from multiple perspectives: teachers completed questionnaires to assess skills in the school context, while parents provided insights from the home context. Given the wide range of participant ages and the variability in symptom severity, this design was particularly suitable, as it allowed us to capture diverse learning trajectories and reflect the children’s progress through gameplay. The study was conducted across school settings, including both mainstream and special education centers, to gather data from multiple perspectives: special education teachers, therapists, and parents. This section details the methodology used, outlining the participant recruitment and study procedures, followed by a description of the outcome measures employed to assess the game’s impact on ADL skills.

Participants and study procedures

Children with autism aged 6 to 15 years who could sustain attention for at least 15 minutes were eligible to participate. Exclusion criteria included attending fewer than two intervention sessions per week or inability to complete the four-week experimental period. Participation was voluntary, with the option to withdraw at any time. Homeroom teachers and parents of the participating children were also included to provide insights into the children’s learning progress. Ethical approval for this study was obtained from the Institutional Review Board at King Mongkut’s University of Technology Thonburi (Approval No. KMUTT-IRB-COA-2021-015). The experimental procedure (Fig. 7) followed these steps:

  1. 1.

    Professionals who work with children with autism in educational settings (schools and special education centers), including occupational therapists and special needs education teachers, were invited to participate as teacher participants. They provided informed consent before receiving detailed information about the study and were responsible for informing the parents of eligible student under their supervision about the research study.

  2. 2.

    Parents who consented to their child’s participation in the study completed and signed both the consent form and the pre-assessment forms, including the Autism Treatment Evaluation Checklist (ATEC) and ADL assessment.

  3. 3.

    Teacher participants also completed the same pre-assessment forms to provide their perspective on the children’s baseline ADL skills.

  4. 4.

    Teacher participants supervised the children’s engagement with the game intervention for four weeks, ensuring a minimum of two sessions per week, with each session lasting at least 20 minutes.

  5. 5.

    Following the four-week intervention, both teacher participants and parents completed the post-assessment forms. Additionally, teacher participants provided detailed insights through structured interviews conducted either online or in person. Additionally, teacher participants participated in structured interviews, conducted either online or in person, to provide qualitative data on the children’s learning progress and their experience with the game.

Fig. 7
figure 7

The quasi-experimental design in this study.

Full size image

Outcome measures and analytical approach

To evaluate the game’s potential to enhance ADL skills in children with autism and the effectiveness of the fuzzy-logic based decision-making system in personalizing learning, we employed a multi-method data collection strategy encompassing both paper-based assessments and in-game data. This approach allowed us to address the research objectives, including assessing skill generalization and the feasibility of the game and its underlying system as an innovative educational tool. The following data collection tools were utilized:

  1. 1.

    Autism Treatment Evaluation Checklist (ATEC): Thai Version – This assessment provided insights into participants’ characteristics by evaluating baseline behaviors related to autism symptoms that could influence their learning abilities across four domains: speech/language and communication, sociability, sensory and cognitive awareness, and physical/health behavior. Domain-specific and overall scores were used to classify participants into mild, moderate, or severe autism symptom severity levels, informing the understanding of potential influences on learning abilities.

  2. 2.

    ADL Skills Assessment Form: This assessment was specifically designed to evaluate participants’ proficiency in performing routine daily living skills. It was developed collaboratively by professionals with extensive experience in providing interventions for children with autism, including an occupational therapist, a developmental psychologist, and a special needs educator. The assessment was aligned with the six learning contexts featured in the game: Home (maximum score = 48), School (48), Canteen (40), Transportation (40), Playground (40), and Shopping (40), ensuring that the game effectively targeted ADL skill development. Skills were rated on a five-level supportive scale, ranging from dependence to supervision/independence: dependence, maximum assistance, moderate assistance, minimal assistance, and supervision/independence. Teacher participants assessed skills in the School, Canteen, Playground, and Shopping contexts, while parent participants assessed skills in the Home, Playground, and Shopping contexts. This assessment provided insights into how children with autism applied the skills learned through the game to real-life situations. The gain scores were calculated to determine the impact of the intervention on ADL skills in real-life applications into positive improvement, negative improvement, or no improvement.

  3. 3.

    Learning Behavior Observation Form: This form was developed to document participants’ learning behaviors during each game interaction session, as rated by the teachers. It assesses essential learning skills such as attention, motivation, and engagement. These skills were rated on a support needs scale (maximum support, moderate support, minimum or no support), providing data on factors critical for goal-directed behavior and learning achievement.

  4. 4.

    Structured Interview: Structured interviews were conducted with teachers post-intervention to gather qualitative insights into children’s learning behaviors, particularly their generalization skills across different learning contexts and activities.

  5. 5.

    Game-based Data Collection: We designed parameters to evaluate player performance and reflect their learning abilities. The game automatically collected time-series data on player performance parameters, including accuracy, reaction and completion time, and earned star. Additionally, data on the fuzzy system’s outputs, such as adjustments to the number of learning modules, were recorded to track the system’s adaptive behavior and participants’ learning trajectories over time.