Introduction

The global agricultural sector faces escalating demands to enhance productivity while minimizing environmental degradation and resource depletion1. In arid and semi-arid regions, date palm cultivation plays a vital socio-economic role serving as a critical source of nutrition, employment, and cultural heritage. However, conventional farming methods largely dependent on manual inspection and farmers’ experiential knowledge are increasingly inadequate to address modern challenges such as climate variability, water scarcity, and soil degradation2. These constraints underscore the need for intelligent, data-driven frameworks that promote precision management and foster long-term sustainability in farming systems.

On a global scale, the date palm industry demonstrates robust growth. The market value is projected to expand from USD 8.3 billion in 2021 to USD 14.2 billion by 2034, indicating a steady upward trend as shown in Fig. 13. Fresh dates account for approximately 45% of the total market, followed by dried dates and syrup-based products. In addition, emerging value-added derivatives such as date powder and paste are gaining traction, driven by increasing consumer preference for natural sweeteners and functional foods4. This sector’s continued growth underscores the rising global demand and diversification of date-based commodities.

Fig. 1
figure 1

Global date palm market size, by product type (2021–2034).

Full size image

The domestic landscape in Saudi Arabia mirrors this global trend. As depicted in Fig. 2, the Kingdom’s date palm export value rose from SAR 0.58 billion in 2016 to SAR 1.70 billion in 2024, achieving a compound annual growth rate (CAGR) of approximately 12.7%5. This acceleration reflects enhanced post-harvest technologies, government-led export initiatives, and a growing emphasis on sustainability within the Saudi Vision 2030 framework. Collectively, these figures reinforce the importance of modernizing date palm cultivation to sustain economic competitiveness while preserving natural resources.

Fig. 2
figure 2

Saudi arabia date palm export value (2016–2024).

Full size image

Contemporary advances in Artificial Intelligence (AI), Machine Learning (ML), and the Internet of Things (IoT) have modernized agricultural systems from manual operations into intelligent, data-driven ecosystems6. By integrating sensor networks, cloud platforms, and predictive algorithms, these technologies enable continuous monitoring of soil, water, and crop parameters turning raw environmental data into actionable insights. Such innovations underpin precision irrigation, automated nutrient management, and early disease detection, thereby enhancing productivity while reducing water consumption and labor requirements.

In the domain of date palm cultivation, the use of AI-driven analytics enables more precise predictions of palm health and yield more accurately. Key biometric parameters such as palm height, trunk diameter, and leaf number, when fused with environmental factors like soil moisture, temperature, and humidity, provide rich datasets for advanced modeling7. ML algorithms including Random Forest (RF), Gradient Boosting Machine (GBM), Artificial Neural Network (ANN), and Support Vector Machine (SVM) have demonstrated efficacy for classifying crop health and forecasting yields under variable field conditions.

Nevertheless, extant studies are often limited by limited, image-centric datasets and lack comprehensive frameworks that integrate multimodal (biometric + environmental) data. Furthermore, very few initiatives within Saudi Arabia have effectively merged AI and IoT for real-time, closed-loop agricultural management. These gaps limit scalability and hinder the full realization of smart farming in arid environments.

To tackle these challenges, this study proposes an AI-enabled Smart Farming Framework specifically tailored for sustainable date palm cultivation in arid regions. The framework unifies biometric and climatic features to enable real-time monitoring, predictive diagnostics, and automated irrigation control. The objectives of this study are delineated as follows:

  1. 1.

    To perform comprehensive data preprocessing and exploratory analysis on the structured palm dataset to identify key relationships between biometric and environmental features.

  2. 2.

    To develop and optimize multiple ML models Random Forest (RF), Gradient Boosting (GBM), Artificial Neural Network (ANN), and Support Vector Machine (SVM) for palm health classification and yield prediction.

  3. 3.

    To design a scalable AI–IoT architecture that supports real-time monitoring, predictive analytics, and adaptive irrigation management tailored to arid-region agriculture.

  4. 4.

    To evaluate model performance using standard classification metrics (accuracy, precision, recall, F1-score) and assess implications for sustainable agriculture aligned with Saudi Vision 2030 objectives.

This study addresses key challenges in smart agriculture and environmental monitoring by integrating AI-driven analytics with IoT-based sensing for continuous observation of soil and microclimatic conditions. The proposed framework is particularly relevant for arid and semi-arid regions, where real-time environmental monitoring and data-driven decision support are essential for sustainable crop management. The remainder of this paper is organized as follows: Section II provides a review of related work on AI and IoT applications in smart agriculture; Section III presents the dataset and methodology; Section IV discusses experimental results and performance evaluation; and Section V concludes with the research findings and highlights potential future research directions.

Literature review

Saudi Arabia, located within an arid and semi-arid climatic zone, faces persistent challenges in achieving sustainable agriculture due to water scarcity, extreme temperatures, and soil salinity. Among its major crops, date palm holds substantial economic, ecological, and cultural significance, contributing directly to national food security and agricultural GDP8. With over 30 million palm trees cultivated across major regions such as Al-Qassim, Al-Ahsa, and Madinah, the Kingdom ranks among the top global producers of dates. However, conventional farming practices largely dependent on manual irrigation scheduling, visual inspection, and farmer experience often result in suboptimal water utilization and inconsistent crop yields9. These inefficiencies underscore the urgent need for intelligent, data-driven solutions tailored to arid agricultural systems.

Recent advancements in Artificial Intelligence (AI), Machine Learning (ML), and the Internet of Things (IoT) have transformed conventional agricultural approaches into technology-driven smart farming ecosystems10. IoT-based systems enable real-time monitoring of soil, humidity, and temperature, while AI algorithms convert this data into actionable insights for optimized decision-making11,12. Through predictive analytics and automation, AI supports precise irrigation control, early disease detection, and yield forecasting, thereby enhancing agricultural productivity and sustainability13. Several pilot projects in Saudi Arabia have already demonstrated the feasibility of AI–IoT integration for smart irrigation, where soil and climate sensors transmit continuous data to predictive models that dynamically regulate water allocation14.

Moreover, deep learning models, particularly convolutional neural networks (CNNs), have achieved remarkable success in visual applications such as fruit grading and palm disease detection15. Traditional ML methods, including Random Forest (RF) and Support Vector Machine (SVM), remain effective for structured agricultural datasets due to their interpretability, scalability, and robustness under noisy environmental conditions16. Despite these promising developments, most existing studies are constrained by limited datasets often focusing exclusively on image-based or single-variable analyses without integrating multimodal environmental and biometric parameters17. This gap restricts model generalization and limits real-world applicability, especially for date palm cultivation, where both climatic and physiological factors play crucial roles in determining crop health and yield.

Table 1 summarizes the current state of AI- and IoT-based agricultural research in Saudi Arabia, identifying recurring limitations such as the lack of open-access multimodal datasets, inadequate AI–IoT integration at the farm level, and insufficient validation of predictive frameworks under field conditions. Although several studies have demonstrated the individual potential of IoT sensors, UAV-based monitoring, and ML models, comprehensive frameworks that combine these technologies for adaptive, real-time decision-making remain scarce.

Table 1 Current state of AI- and IoT-based agricultural research in Saudi Arabia.
Full size table

In response to these research gaps, the present study proposes an integrated AI-enabled smart farming framework specifically designed for date palm cultivation in Saudi Arabia’s arid environment. By combining biometric indicators (e.g., palm height, trunk diameter, and leaf number) with environmental parameters (e.g., soil moisture, humidity, and temperature), the framework aims to establish a robust data-driven decision support system for real-time monitoring, predictive analytics, and sustainable resource management. This integrative approach advances the practical realization of precision agriculture in alignment with Saudi Vision 2030 objectives, promoting agricultural resilience and digital transformation.

Materials and methods

Dataset description

The Structured Palm Dataset employed in this study consists of 500 real-world records collected from date palms cultivated in arid regions of Saudi Arabia. Each record captures eight critical features Palm Height, Trunk Diameter, Leaf Number, Soil Moisture, Temperature, Humidity, Variety, and Health Status which together provide a comprehensive representation of both the physiological characteristics of the palms and the environmental factors affecting their growth and productivity. A stratified random sampling approach was employed during data collection to ensure proportional representation of different palm varieties and health status classes, enabling balanced class distribution and robust supervised learning.

This dataset constitutes a fundamental basis for developing data-driven analytics in smart farming applications throughout Saudi Arabia, where date palm is a strategic crop for food security and economic development. The dataset provides a comprehensive depiction of plant–environment interactions, facilitating predictive modeling and intelligent decision-making for irrigation management, disease diagnosis, and yield estimation. Table 2 presents a sample of the structured dataset, while Table 3 provides detailed descriptions of each feature, including data type, unit of measurement, and contextual significance. The integration of biometric features (palm height, trunk diameter, and leaf number), environmental parameters (soil moisture, temperature, and humidity), and categorical attributes (variety and health status) enables comprehensive modeling of palm growth, microclimatic stress, and disease conditions, thereby enhancing classification accuracy and decision reliability.

Table 2 Structured palm dataset sample.
Full size table
Table 3 Description of dataset features.
Full size table

Data preprocessing

To ensure data reliability and consistency, it was subjected to a rigorous preprocessing pipeline prior to model imputation27. Missing values were imputed using median substitution for numerical variables and mode replacement for categorical features. Outliers particularly in trunk diameter and soil moisture were detected through the Interquartile Range (IQR) method and corrected using normalization techniques to preserve statistical stability.

Categorical variables such as Variety and Health Status were numerically encoded using label encoding, facilitating seamless integration with machine learning algorithms. All numerical features were standardized through z-score normalization to ensure uniform feature contribution during model training28. The processed dataset was subsequently partitioned into 80% training and 20% testing subsets to enable robust validation and prevent model overfitting.

Model development

Four supervised learning algorithms Random Forest (RF), Support Vector Machine (SVM), Gradient Boosting Machine (GBM), and Artificial Neural Network (ANN) were implemented and optimized for classifying palm health and predicting crop variety29. The Random Forest algorithm was selected for its interpretability and capability to handle heterogeneous feature types, while SVM with a radial basis function kernel was employed to capture nonlinear relationships within the data. GBM was utilized to enhance model accuracy through iterative boosting, and ANN was configured with two hidden layers to capture complex nonlinear dependencies between biometric and environmental parameters.

Model hyperparameters were tuned via a grid search combined with five-fold cross-validation, ensuring an optimal balance between computational efficiency and prediction accuracy30. The resulting optimized models provided a robust comparative foundation for evaluating machine learning performance in arid agricultural settings.

Model evaluation metrics

The performance of the machine learning models was evaluated using standard classification metrics: Accuracy, Precision, Recall, and F1-Score31. These metrics are mathematically expressed as follows:

$$\:Accuracy=\frac{TP+TN}{TP+TN+FP+FN}$$
(1)
$$\:Precision=\frac{TP}{TP+FP}$$
(2)
$$\:Recall=\frac{TP}{TP+FN}$$
(3)
$$\:F1\:Score=2\times\:\frac{Precision\times\:Recall}{Precision+Recall}$$
(4)

where \(\:TP\), \(\:TN\), \(\:FP\), and \(\:FN\:\)denote true positives, true negatives, false positives, and false negatives, respectively.

Accuracy measures evaluates overall classification correctness, while Precision reflects the reliability of positive predictions. Recall quantifies model sensitivity, and the F1-Score provides a balanced measure between Precision and Recall particularly relevant when dealing with imbalanced datasets. Confusion matrices were also generated for each model to visually assess classification performance and identify misclassification trends.

The coefficient of determination \(\:R^2\) is a statistical measure that indicates how well a model’s predictions match the actual observed data32. It is calculated using the formula:

$$\:{R}^{2}=1-\frac{\sum\:_{i=1}^{n}{({y}_{i}-{\widehat{y}}_{i})}^{2}}{\sum\:_{i=1}^{n}{({y}_{i}-{\stackrel{-}{y}}_{i})}^{2}}$$
(5)

where \(\:{y}_{i}\)​ represents the actual observed values, \(\:{\widehat{y}}_{i}\:\)represents the predicted values from the model, \(\:{\stackrel{-}{y}}_{i}\) is the mean of the observed values, and \(\:n\) is the total number of observations. \(\:R^2\) ranges from 0 to 1, with a value of 1 indicating perfect prediction accuracy and a value of 0 indicating that the model does not explain any variability in the observed data. This metric is widely used in regression analysis to evaluate the goodness of fit of predictive models.

Proposed AI-based smart farming framework

The proposed AI-enabled Smart Farming Framework integrates IoT-enabled sensor networks and machine learning analytics to facilitate sustainable date palm cultivation under arid environmental conditions. The framework consists four core layers: data collection, data processing, intelligent analytics, and decision support, as illustrated in Fig. 3. Figure 4 illustrates the data-driven decision support flowchart for monitoring palm health.

Fig. 3
figure 3

Proposed AI-based smart farming framework.

Full size image
Fig. 4
figure 4

Smart farming data-driven decision support flowchart for date palm monitoring.

Full size image

In the data acquisition layer, biometric and environmental parameters such as soil moisture, temperature, humidity, and palm growth metrics are captured via IoT sensors and transmitted to a cloud-based database enabling real-time monitoring. The IoT sensing subsystem was configured to capture both root-zone and canopy-level microclimatic conditions surrounding each date palm. Soil moisture sensors were installed at a depth of 25–30 cm within the root zone to accurately measure volumetric water content influencing nutrient uptake and irrigation efficiency. Temperature and relative humidity sensors were mounted at approximately 1.5–2.0 m above ground level near the palm canopy to monitor ambient environmental conditions directly affecting transpiration and physiological stress. All sensors were spatially arranged around the palm tree in a radial configuration to ensure uniform microclimate coverage, as illustrated in Fig. 5. Sensor data were transmitted through a low-power wireless communication module to a central gateway and forwarded to the cloud-based analytics platform, as depicted in Fig. 6. Prior to deployment, each sensor was calibrated against standard reference instruments to minimize measurement drift and noise, ensuring high data reliability for machine learning–based analytics.

Fig. 5
figure 5

Sensor setting around the palm tree.

Full size image
Fig. 6
figure 6

Sensor signal flow within the AI–IoT framework.

Full size image

The data preprocessing layer performs cleaning, normalization, and feature engineering to enhance dataset quality and reduce noise. In the intelligent analytics layer, trained ML models, including RF and GBM are deployed for health classification, yield estimation, and environmental prediction. The decision support layer subsequently translates these insights into actionable recommendations, including automated irrigation scheduling and early disease alerts delivered via mobile or web dashboards.

This multi-tiered architecture enables real-time decision-making, predictive diagnostics, and resource-efficient management. Its modular and scalable design aligns with Saudi Vision 2030, supporting the nation’s goals of agricultural innovation, water efficiency, and food security through digital transformation.

Results and discussion

This section provides the analytical findings of the proposed AI-based smart farming framework, including exploratory data analysis, model evaluation, and comparative performance assessment. The discussion further interprets these results in the context of Saudi Arabia’s agricultural transformation goals and sustainability initiatives under Vision 2030.

Data analysis

A comprehensive exploratory data analysis (EDA) was conducted to examine correlations among biometric and environmental attributes affecting palm health and productivity. Figure 7 depicts the correlation heatmap for the key features in the Structured Palm Dataset. A strong positive correlation was detected between palm height and trunk diameter, confirming their joint contribution to yield potential. Moreover, humidity and soil moisture exhibited moderate correlations with palm health status, highlighting their vital role in stress prediction and disease detection within arid environments.

Descriptive statistics analysis revealed a well-balanced dataset, with nearly equal representation of healthy and diseased samples. This balance minimized potential model bias during training and ensured reliable generalization across health classes. The heterogeneous combination of biometric and environmental features created a comprehensive feature space, improving model interpretability and enhancing the robustness of subsequent predictive analytics.

Fig. 7
figure 7

Correlation heatmap among biometric and environmental features in the date palm dataset.

Full size image

Model performance

Four supervised machine learning models Random Forest (RF), Gradient Boosting Machine (GBM), Artificial Neural Network (ANN), and Support Vector Machine (SVM) were trained and evaluated using the preprocessed dataset. Each model was optimized through grid search and five-fold cross-validation to guarantee consistent performance across diverse feature subsets. Table 4 summarizes the training time and hyperparameter tuning configuration for all evaluated machine learning models. Hyperparameters were optimized using grid search with five-fold cross-validation to ensure fair and robust performance comparison. The results indicate that ANN required the longest training time due to iterative learning, while Random Forest and Gradient Boosting achieved high accuracy with moderate computational cost.

Table 4 Training time and hyperparameter tuning configuration of ML models.
Full size table

The comparative performance outcomes are summarized in Table 5. The Random Forest model achieved the highest classification accuracy of 95.3%, with precision and recall values of 94.7% and 96.1%, respectively. This demonstrates its superior ability to capture nonlinear dependencies and handle heterogeneous input variables effectively. The GBM model followed closely, exhibiting high stability and predictive reliability. Conversely, the ANN model achieved good accuracy (92.4%) but required longer training time and extensive hyperparameter tuning. The SVM model, while computationally efficient, exhibited relatively lower accuracy (89.7%) due to its limited capacity to model high-dimensional relationships.

Table 5 Model performance comparison.
Full size table

Of the evaluated algorithms, the Random Forest (RF) model achieved the highest overall accuracy of 95.3% and an F1-Score of 95.4%, confirming its robustness and ability to handle heterogeneous features. The GBM model closely followed, demonstrating stable generalization performance. Although the ANN model showed good predictive ability, it required greater computational time and hyperparameter tuning. Conversely, the SVM model, while efficient for smaller datasets, exhibited reduced accuracy owing to the nonlinear and high-dimensional relationships between variables.

Figure 8 depicts the performance of the Random Forest (RF) model for soil moisture prediction. The model effectively captured soil moisture variability, achieving an R² of 0.982 and an RMSE of 0.54%. Predicted values closely matched actual sensor readings, with the majority of deviations within ± 2%. This strong correlation underscores the model’s ability to generalize across varying soil conditions, encompassing moisture levels from 18.5% to 36.9%. Accurate moisture prediction is crucial for optimizing irrigation schedules, minimizing water wastage, and maintaining ideal soil conditions for date palm cultivation in arid environments33.

Fig. 8
figure 8

Performance of the Random Forest–based soil moisture prediction model.

Full size image

Figure 9 shows the performance of the Random Forest model for soil pH prediction. The model achieved stable performance with an R² of 0.975 and a mean absolute error below 1%. Observed and predicted pH values ranged from 6.6 to 7.6, indicating a neutral to slightly alkaline soil profile suitable for date palms. The RF model’s precision highlights its ability to detect subtle chemical variations influenced by irrigation water quality and fertilizer composition, thereby facilitating timely interventions for nutrient management.

Fig. 9
figure 9

Performance of the random forest–based soil pH prediction model.

Full size image

Figure 10 illustrates the performance of the Random Forest model for soil temperature prediction. The model attained an R² of 0.984 and an RMSE of 0.42 °C. Actual soil temperatures ranged from 30.2 °C to 37.4 °C, and the predicted trends closely followed these measurements. This alignment demonstrates the system’s robustness in capturing microclimatic variations around the palm root zone34. Reliable temperature forecasting supports optimal irrigation timing and helps prevent thermal stress, which is essential for sustaining crop yield and quality under arid desert conditions.

Fig. 10
figure 10

Performance of the random forest–based soil temperature prediction model.

Full size image

Figure 11 presents the deviation analysis between actual and predicted soil parameters. The overall deviation averaged merely 1.3%, validating the accuracy and stability of the AI–IoT framework. The minimal error across all parameters confirms the Random Forest model’s effectiveness in handling multi-sensor data, compensating for potential sensor noise or latency, and providing reliable guidance for automated irrigation and soil management without constant human supervision35.

Fig. 11
figure 11

Deviation analysis between actual and predicted soil parameters.

Full size image

Beyond classification accuracy, the Random Forest model demonstrated strong predictive capability for key environmental features, including soil moisture, soil temperature, and soil pH. As shown in Figs. 8, 9, 10 and 11, the model achieved coefficients of determination exceeding 0.97, indicating accurate estimation of sensor-derived parameters essential for irrigation scheduling and soil management.

Figure 12 presents the Receiver Operating Characteristic (ROC) curves for the Random Forest, Gradient Boosting Machine, Artificial Neural Network, and Support Vector Machine classifiers. The Random Forest model achieved the highest Area Under the Curve (AUC = 0.97), indicating superior discriminative capability. GBM and ANN also demonstrated strong performance with AUC values of 0.95 and 0.94, respectively, while SVM exhibited comparatively lower separability (AUC = 0.91). These results confirm the robustness and reliability of the proposed framework for palm health classification.

Fig. 12
figure 12

Receiver operating characteristic curve for palm health classification models.

Full size image

Figure 12 illustrates the comparison between observed and predicted leaf health classifications across 500 samples. The close alignment between bar heights across the three health categories (healthy, diseased, and nutrient-deficient) indicates strong model accuracy with minimal misclassification. Minor deviations were primarily observed between the diseased and nutrient-deficient categories, emphasizing the model’s high sensitivity for early disease detection in smart farming applications.

Fig. 13
figure 13

Comparison between observed and predicted leaf health classifications using the Random Forest model.

Full size image

Sensitivity and sensor calibration analysis

To assess environmental robustness, a sensitivity analysis of the Random Forest model was performed under varying environmental conditions (± 10% perturbations in temperature, humidity, and soil moisture). As depicted in Fig. 13 maintained stable accuracy with only minimal fluctuations, confirming strong resilience against environmental noise and dynamic field variability. Ensemble averaging within the Random Forest architecture effectively mitigated random perturbations, ensuring consistent predictive stability under real-world conditions. The calibrated sensor configuration ensured stable and accurate data acquisition under arid field conditions, providing reliable inputs for the AI models and contributing to consistent predictive performance.

Figure 14 illustrates the calibration performance of five environmental sensors integrated within the AI–IoT framework. The sensors exhibited strong linearity and high precision with coefficients of determination (R² > 0.97). Among them, the humidity sensor demonstrated the highest calibration accuracy (R² = 0.987), followed by the temperature (R² = 0.984) and wind speed (R² = 0.981) sensors. The volumetric water content and flow rate sensors achieved R² = 0.981 and R² = 0.977, respectively, confirming reliable measurements essential for water management and irrigation optimization. These results substantiate the overall robustness of the environmental sensing subsystem and its suitability for continuous operation in arid field conditions.

Fig. 14
figure 14

Sensitivity analysis of the random forest model’s accuracy under varying environmental conditions.

Full size image
Fig. 15
figure 15

Calibration performance of environmental sensors used in the smart farming framework.

Full size image

Confusion matrix and model validation

Figure 15, 16 depicts the confusion matrix for the Random Forest classifier. The results exhibit minimal misclassification between healthy and diseased classes, confirming high discriminative performance. The model achieved balanced precision (94.7%) and recall (96.1%), indicating effective identification of both healthy and infected palms. The low off-diagonal values highlight the model’s strong generalization capability and substantiate its applicability in real-world farm monitoring scenarios.

Overall, the Random Forest model exhibited strong predictive consistency and interpretability, making it well-suited for integration with IoT- enabled systems. Its ensemble nature ensures stability even under fluctuating environmental conditions, addressing key challenges in agricultural automation such as data variability, constrained sample size, and sensor drift.

Fig. 16
figure 16

Confusion matrix for the classification performance of the Random Forest model.

Full size image

Comparative discussion and implications

Comparative analysis against existing AI-based agricultural studies revealed that the proposed framework attained superior accuracy and stability due to the integration of multimodal features combining both biometric and environmental data. From an environmental monitoring perspective, the proposed AI–IoT framework enables continuous assessment of soil moisture, temperature, and humidity, supporting timely detection of stress conditions and efficient resource utilization. This integration demonstrates the practical relevance of the framework for precision agriculture and environmental sustainability, especially in water-scarce regions. This fusion improved model resilience against climatic fluctuations, a major limitation in prior single-domain approaches focusing solely on image or environmental inputs. The strong predictive performance validates the framework’s potential as a decision support tool for precision irrigation, early disease detection, and yield optimization in arid-zone agriculture. When deployed at scale, the AI–IoT integration could significantly reduce manual labor dependency, optimize water use, and increase productivity per hectare.

As shown in Table 6, the proposed AI–IoT framework outperforms existing approaches in terms of classification accuracy. Unlike most state-of-the-art studies that rely solely on image-based or limited environmental features, this work integrates multimodal biometric, environmental, and categorical data, resulting in improved robustness, generalization, and predictive reliability.

Table 6 Comparison with state-of-the-art AI-based smart agriculture studies.
Full size table

Moreover, the studies align with Saudi Vision 2030, underpinning the national objectives of digital transformation and sustainable resource management. The presented framework offers a scalable and replicable framework that can be adapted for other climate-sensitive crops, enhancing regional food security and agricultural resilience.

Conclusion

This study presents an AI–IoT- integrated smart farming framework to enhance the sustainability and productivity of date palm farming in Saudi Arabia’s arid regions. By integrating biometric indicators (palm height, trunk diameter, leaf number) with environmental parameters (soil moisture, temperature, humidity), the framework enables data-driven management of palm health and yield.

Four supervised machine learning models Random Forest (RF), Gradient Boosting Machine (GBM), Artificial Neural Network (ANN), and Support Vector Machine (SVM) were evaluated for palm health classification. The Random Forest model achieved the highest accuracy of 95.3% and demonstrated strong generalization in predicting environmental variables (R² > 0.97), supporting irrigation scheduling, disease detection, and yield forecasting.

Through the integration of AI-driven analytics with IoT sensor networks, the framework provides real-time monitoring, predictive diagnostics, and automated control of irrigation and environmental systems, advancing precision agriculture in arid zones. In alignment with Saudi Vision 2030, it promotes smart agriculture, water efficiency, and food security, offering a replicable model for other climate-sensitive crops in similar regions.

Future work will focus on real-time deployment of IoT infrastructures, integration of remote sensing, edge computing, and advanced deep learning (CNNs, LSTMs), alongside a cloud-based farmer dashboard for predictive alerts, enabling large-scale adoption of AI-driven precision farming.