A Robust Method for Diabetes Detection Using Deep Learning with Convolutional LSTM Networks

Abstract

A common metabolic disease that can impact billions of individuals worldwide, diabetes mellitus is a chronic illness. Diabetes complications can be adequately prevented with ongoing monitoring and early identification. In this study, we use ConvLSTM networks and deep learning techniques to develop a reliable method for diabetes detection. Because of its spatiotemporal properties, ConvLSTM, which combines Convolutional Neural Networks (CNN) with Long Short-Term Memory (LSTM), is well-suited for processing medical data. Compared to approaches created before it and other deep learning models, our approach demonstrated to have superior accuracy and efficiency. In benchmark datasets, our algorithm proved efficacy in early diabetes identification and categorization.

Introduction

Diabetes is a growing global health issue, expected to affect over 643 million people by 2030. Early and accurate diagnosis is crucial to prevent serious complications. Traditional methods rely on clinical knowledge and glucose tests but often miss complex patterns in data.

To address this, the study proposes an automated diabetes diagnosis system using Convolutional Long Short-Term Memory (ConvLSTM) networks. ConvLSTM combines the spatial strengths of Convolutional Neural Networks (CNNs) with the temporal modeling capabilities of Long Short-Term Memory (LSTM) networks, making it ideal for dynamic, medical time-series data.

Objectives:

1. Develop a deep learning model (ConvLSTM) to detect spatiotemporal patterns in clinical data.

2. Improve diagnostic accuracy over conventional ML and deep learning models.

3. Use datasets like the Pima Indians Diabetes Database (PIDD) for training and evaluation.

4. Compare the performance against CNNs, LSTMs, and traditional ML models.

5. Support early detection for personalized treatment planning.

Methodology:

• Dataset: PIDD (768 samples, 9 features like glucose, insulin, BMI, age).

• Preprocessing: Handles missing values, normalizes features, and structures data for ConvLSTM input.

• Model: ConvLSTM architecture includes convolutional and recurrent layers for spatiotemporal learning, with dropout for regularization.

• Training: Uses Adam optimizer, binary cross-entropy loss, early stopping, and learning rate scheduling.

Mathematical Model (ConvLSTM):

ConvLSTM replaces standard LSTM matrix operations with convolutions, preserving spatial features over time—ideal for health metrics that vary both spatially and temporally (e.g., glucose levels across time).

Evaluation:

Performance is measured using:

• Accuracy

• Precision

• Recall

• F1-score

• ROC-AUC

Related Work:

Many prior studies used algorithms like KNN, SVM, Decision Trees, ANN, CNN, and LSTM. While these show promise, they often lack in handling both spatial and temporal dependencies, or suffer from dataset imbalance and generalization issues. ConvLSTM addresses these gaps effectively.

Conclusion

This research employs ConvLSTM networks in recommending a dependable approach to diabetes diagnosis. The model is more efficient than other deep learning models and traditional models since it is able to extract spatiotemporal information. The future of research will encompass wearable device integration and real-time monitoring systems.

References

[1] International Federation of Diabetes (2021). The 10th edition of the IDF Diabetes Atlas. [2] Kidd, B. A., Dudley, J. T., Li, L., and Miotto, R. (2016). Deep Patient: An unsupervised model that uses electronic medical information to forecast patients\' futures. 6, 26094, Scientific Reports. [3] Sisodia, D., and Sisodia, D. S. (2018). diabetes prediction using classification algorithms. 1578–1585 in Procedia Computer Science, 132. [4] Andreu-Perez, J., Lo, B., Berthelot, M., Ravi, D., Wong, C., Deligianni, F., & Yang, G. Z. (2017). Health informatics using deep learning. Biomedical and Health Informatics, IEEE Journal, 21(1), 4–21. [5] Wang, H., Yeung, D. Y., Wong, W. K., Woo, W. C., Shi, X., Chen, Z., & Woo in 2015. A machine learning approach to precipitation nowcasting is the convolutional LSTM network. Neural Information Processing Systems Advances, 28. [6] Graff, C., and Dua, D. (2017). UCI Repository for Machine Learning. School of Information and Computer Sciences, University of California, Irvine. [7] Nancy, A. O. V., Arya, M. S., and Ahamed, B. S. (2022). Predicting Diabetes Mellitus Using Machine Learning Classifiers with Feature Augmentation and Oversampling. developments in computer-human interaction. [8] Campanile, L., Mainenti, G., Ricciardi, C., Marulli, F., & Valente, A. S. (2020). Machine Learning Approaches for Categorization of Diabetes: Perspectives on the Revising Artificial Intelligence Methods. IoTBDS, pp. 533-540. [9] Ksantini, M., Alsaif, H., Kahouli, O., Ellouze, A., Aloui, A., &Kahouli, B. (2022). Diabetes Diagnosis Based on Artificial Intelligence and Belief Functions Theory. Symmetry, 14(10), 2197. [10] In May 2021, Llaha, O., and Rista, A. Diabetes prediction and detection using machine learning. On pages 94–102 of RTA-CSIT. [11] Zakir, S. A., Raheem, M. A., Ehetesham, S., and Subhani, M. F. A. AI Algorithm System for Diabetes Prediction Using Progressive Web App and IBM Cloud. [12] Taleb, S. F., Tibhirt, A., Chenikha, K., Sadoun, R., Benalia, N. E. H., &Bentrah, A. (2022). A machine learning-based diabetes prediction tool using System on Chip. Evolution of Artificial Intelligence, 69-86. [13] Soni, S., Vakil, V., Pachchigar, S., and Chavda, C. (2021). Comparison of predictions of various machine learning algorithms for prognosis of early-stage diabetes. 2111.09939 is the arXiv preprint. [14] Kumar, R. U., and P. B. K. Chowdary (2021). A Successful Approach for Diabetes Diagnosis Using Convolutional LSTM Network-Based Deep Learning Techniques. Advanced Computer Science and Applications International Journal, 12(4). [15] Abu-Naser, S. S., and El_Jerjawi, N. S. (2018). Predicting diabetes using artificial neural networks. [16] Kumar, R. U., and P. B. K. Chowdary (2021). An optimal approach for the detection of diabetes using convolutional LSTM network-based deep learning techniques. 12(4), 519-525, International Journal of Advanced Computer Science and Applications. [17] Kaggle, data set “Diabetes Dataset (kaggle.com)”

Copyright

Copyright © 2025 Abhishek Shahi, V. K. Sharma, Anshu Kumar Dwivedi, Amitabh Wahi. 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.