An Ensemble Deep Learning Framework for Early Diabetes Prediction Using Clinical and Lifestyle Features

Abstract: Early detection of diabetes is crucial for effective management and prevention of complications. This research presents a robust deep learning framework for predicting diabetes using clinical and lifestyle features. A fully connected neural network model with batch normalization, dropout layers, and residual connections was designed to handle class imbalance and improve generalization. The model was trained on a comprehensive dataset of 100,000 patient records and evaluated using accuracy, precision, recall, and F1-score metrics. Experimental results demonstrate that the proposed approach achieves a test accuracy of 97.13%, outperforming conventional machine learning models and recent state-of-the-art methods. Confusion matrix and classification reports confirm high predictive performance for both positive and negative classes. This framework provides a scalable, interpretable, and efficient solution for early diabetes screening in healthcare systems.

Keywords: Deep Learning, diabetes prediction, early detection, Heart rate variability, ECG, CNN, LSTM