AI-Powered Hybrid Deep Learning Model for Early Diagnosis of Type-2 Diabetes Mellitus

Abstract

Type 2 Diabetes Mellitus (T2DM) is a rapidly growing global health concern characterized by long-term complications and high mortality rates. Early diagnosis is essential to reduce risks and improve patient outcomes. This study proposes a hybrid deep learning model that integrates a Multi-Layer Perceptron (MLP) and Long Short-Term Memory (LSTM) network for accurate and reliable diabetes prediction. The model is trained on the Pima Indians Diabetes Dataset after systematic preprocessing and normalization. To address the interpretability challenges of deep learning, SHapley Additive exPlanations (SHAP) is incorporated to provide transparent insights into model predictions. Experimental results demonstrate that the hybrid model achieves superior performance compared to individual models, with improved accuracy, recall, and F1-score. The integration of explainable AI enhances trust and usability in clinical environments, making the proposed system a promising solution for early diabetes risk assessment.

Introduction

Type 2 Diabetes Mellitus is a chronic metabolic disease caused by insulin resistance and poor glucose regulation. With the increasing global prevalence of diabetes, there is a growing need for accurate early prediction systems. Traditional machine learning models such as logistic regression, decision trees, and support vector machines provide moderate prediction accuracy but struggle to capture complex nonlinear relationships in medical data. Deep learning methods improve prediction performance but often lack interpretability, limiting their use in healthcare.

This research proposes a hybrid deep learning framework that combines Multi-Layer Perceptron (MLP) and Long Short-Term Memory (LSTM) models to improve diabetes prediction. The MLP captures nonlinear feature relationships in tabular clinical data, while the LSTM identifies sequential dependencies among features. The study uses the Pima Indians Diabetes Dataset containing 768 patient records with attributes such as glucose level, BMI, insulin, blood pressure, and age.

Data preprocessing included handling missing values through median imputation, feature normalization, dataset splitting, and reshaping data for LSTM input. To improve transparency and trustworthiness, SHAP explainability was integrated to identify the contribution of each feature to predictions.

Results showed that the hybrid model achieved the best performance with 91% accuracy, outperforming standalone MLP and LSTM models. SHAP analysis identified glucose as the most important predictor, followed by BMI and insulin levels, while age had a moderate influence. The findings demonstrate that the hybrid approach improves prediction accuracy while maintaining interpretability, making it more suitable for clinical applications.

Conclusion

This study presents a hybrid deep learning framework for early diagnosis of Type 2 Diabetes Mellitus. By integrating MLP and LSTM architectures with SHAP-based explainability, the model achieves high predictive accuracy and transparency. The proposed approach addresses key limitations of existing models by balancing performance and interpretability. It provides a practical solution for healthcare systems aiming to implement AI-driven decision support tools. Future work will focus on incorporating larger and more diverse datasets, real-time monitoring systems, and advanced explainability techniques to further enhance model performance and usability.

References

[1] International Diabetes Federation. (2023). IDF Diabetes Atlas (10th ed.). Brussels, Belgium: IDF. [2] Kumar, R., Arora, R., & Shukla, M. (2024). Comparative study of machine learning models for diabetes prediction. Biomedical Signal Processing and Control, 93, 106048. https://doi.org/10.1016/j.bspc.2024.106048 [3] Ahsan, G. M., & Rahman, M. (2024). Hybrid AI in healthcare: Advances in deep neural networks for diabetes prediction. IEEE Access, 12, 15876-15888. https://doi.org/10.1109/ACCESS.2024.3352187 [4] Gupta, D., Sharma, S., & Singh, H. (2023). Enhancing medical diagnosis using deep neural networks: A case study on diabetes. Health Informatics Journal, 29(4), 14604582231122389. https://doi.org/10.1177/14604582231122389 [5] Huang, Y., Zhang, Y., & Zhao, X. (2025). Explainable AI in clinical diagnostics: A review of interpretable deep learning methods. Artificial Intelligence in Medicine, 154, 102955. https://doi.org/10.1016/j.artmed.2025.102955 [6] Qureshi, M. A., et al. (2024). Evaluation of classical machine learning algorithms for diabetes prediction. Informatics in Medicine Unlocked, 39, 101313. https://doi.org/10.1016/j.imu.2024.101313 [7] Jha, K., & Ghosh, D. (2023). Temporal modeling of medical attributes for diabetes detection using recurrent neural networks. Scientific Reports, 13(1), 15522. https://doi.org/10.1038/s41598-023-42612-y [8] Singh, P., & Mishra, V. (2024). The trust gap in black-box medical AI: Explainability as a solution. Frontiers in Artificial Intelligence, 7, 1452321. https://doi.org/10.3389/frai.2024.1452321 [9] Wang, H., & Li, Z. (2025). Global and local explainability for healthcare AI systems using SHAP and integrated gradients. IEEE Transactions on Computational Social Systems, 12(2), 345-358. https://doi.org/10.1109/TCSS.2025.3344558 [10] Allani, U. (2025). Interactive diabetes risk prediction using explainable machine learning: a Dash-based approach with SHAP, LIME, and comorbidity insights. arXiv preprint. https://arxiv.org/abs/2505.05683 [11] Alvarado, M. E., & Kumar, P. (2023). Machine learning in metabolic disorder diagnostics: A comparative review. Computers in Biology and Medicine, 168, 107580. https://doi.org/10.1016/j.compbiomed.2023.107580 [12] Qureshi, M. A., et al. (2024). Evaluation of classical machine learning algorithms for diabetes prediction. Informatics in Medicine Unlocked, 39, 101313. https://doi.org/10.1016/j.imu.2024.101313

Copyright

Copyright © 2026 Maharshikumar K. Vyas. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.