Diabetes prediction: A two-stage framework for balancing accuracy and interpretability using machine learning and deep learning

Diabetes, also known as diabetes mellitus, is a chronic metabolic illness that affects millions of people around the world. Early detection is critical to avoiding serious complications. This study shows a two-stage methodology for comparing Machine Learning (ML) and Deep Learning (DL) models for diabetes prediction. In the first stage, four machine learning models (Random Forest, SVM, XGBoost, and Decision Tree) are evaluated for interpretability and computational efficiency. The second stage compares the best-performing ML model (Random Forest) to two DL models (LSTM and DNN). The results provide that Random Forest has a ROC-AUC of 0.815, beating DL models in interpretability, while LSTM achieves the best accuracy (0.710). The study focuses on the trade-offs between accuracy, interpretability, and computing cost, providing useful insights for healthcare applications.