Hybrid Deep Learning Approach for Accurate Early-Stage Diabetes Risk Prediction

Timely identification of diabetes is essential for preventing serious complications and enhance patient outcomes. This paper introduces a Hybrid Deep Learning Approach that integrates Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks for precise early-stage diabetes risk prediction. The proposed model utilizes CNN for automatic feature extraction from clinical and demographic data, while LSTM identifies temporal patterns and dependencies within patient records. The system was trained and assessed using the Early-Stage Diabetes Risk Prediction Dataset from Kaggle, comprising 520 cases and predictive attributes. Experimental findings indicate that the hybrid CNN-LSTM model attains an accuracy of 99.42 % improving existing machine learning (ML) models. The model has strong generalization, with negligible overfitting in the training-validation performance curves. The results demonstrate that the integration of CNN and LSTM successfully captures spatial and sequential patterns in medical data, facilitating accurate early prediction of diabetes risk. This technology aids healthcare workers in proactive intervention strategies, hence enhancing illness management and prevention.