Table 1 Summary of related works.

(Hennebelle et al. 2024)²³

DP via secured IOT with BC monitoring

Diabetes mellitus prediction model (DMPM) and Random Forest (RF)

Ensured data privacy by storing the diagnosed result in the BC

As the traditional consensus protocol and authentication scheme were utilized, the security was reduced

(R. Huang et al. 2022)²⁵

Diabetes management model

Delay-aware resource allocation (RA) and deep forest algorithm (DFA)

Effectively transmitted the data and attained high diabetes classification accuracy

The network collision was not avoided, leading to higher access delays

(Prabhakar et al. 2024)²⁶

Type 2 DP and diet plan recommendation

Decision Tree, support vector machine (SVM), and artificial neural network (ANN)

Attained higher scalability and prediction accuracy

Required higher memory and was computationally complex

(Razfar et al. 2023)²⁴

Smart post-stroke assessment in the wearable device

Multi-Level Meta Learner (MLML) ensemble classifier

Computational efficiency was better

As only the upper body movement was concentrated, the misclassification rate was higher

(Mahesh et al. 2022)²⁷

DP model

Bayesian networks and radial basis functions

DP accuracy was higher

The blending of base learners caused higher processing time

(Allen et al. 2021)²⁸

Forecasting disease progression in stroke patients

Digital twin and variational-autoencoder

Improved clinical decision-making

The dataset was small and the generalization was reduced

(Annamalai & Nedunchelian 2021)²²

Diabetes mellitus prediction and severity estimation

Switching midvalue-centric morphological filter (SMVMF) and optimal weighted Deep artificial neural network (OWDANN)

Diabetes was classified with high efficiency

The severity level estimation was poor due to the unstable score values

(Lu & Wang 2022)²⁹

Ischemic stroke prediction

Back propagation (BP) network and multivariate logistic regression (MLR)

The ischemic stroke was detected with high accuracy

As the number of iterations increased, the training time also increased

(Elbagoury et al. 2023)³⁰

Stroke prediction via smart hospital platform

Convolutional neural network (CNN), group handling method (GMDH), and long short term memory (LSTM)

Obtained higher stroke prediction accuracy

The variation of the signal channel led to more memory usage and computational complexity

(Krishnamoorthi et al. 2023)³¹

Diabetes healthcare disease prediction

Logistic regression (LR)

The error rate was minimized

The hyper-parameter selection was degraded due to the random sample distribution

(Rastogi & Bansal 2023)³²

DP model

KDD, RF, SV, logistic regression, and Naive Bayes

Improved accuracy in DP

Missing values in the data reduced the performance

(Zhou et al. 2023)³³

Diabetes identification model

Boruta, K-Means +  + , and ensemble classifier

DP precision was higher

The absence of data imbalance led to improper classified output

(Al Reshan et al. 2024)³⁴

DL clinical decision-making regarding DP

ETC, ANN, CNN, and LSTM

Higher accuracy in DP

The model was complex and slowed the DP

(Sai et al. 2023)³⁵

Prediction of diabetes

LGB and adaptive boosting

Attained higher processing time

Reduction in proper decision-making

(Chee et al. 2024)³⁶

Detection of diabetes

CNN + LSTM

Lowered computational complexity

The misclassification rate was higher