Diabetes Detection with Hybrid Deep Learning Models

Diabetes, a chronic condition characterized by persistent hyperglycemia, poses growing global health challenges. While machine learning (ML) and deep learning (DL) models show diagnostic promise, their standalone implementations face limitations—interpretability challenges in DL and limited feature extraction in ML. To address this, we propose two innovative frameworks: (1) hybrid ML-DL architectures (CNN-RF, CNN-KNN) and (2) a deep learning hybrid (CNN-LSTM), evaluated on the PIMA Indians dataset. Our CNN-LSTM combines 1D convolutional layers (64/128 filters) with a 50-unit LSTM, achieving 84% accuracy—significantly outperforming standalone LSTM (78%) and RF (76%) (p ¡ 0.01). It also demonstrated clinically relevant metrics: precision = 0.88, F1-score = 0.89, and recall = 0.88, reducing false negatives by 15% compared to benchmarks. These results highlight the transformative potential of hybrid architectures, particularly CNN-LSTM, for accurate diabetes diagnosis. Future work will focus on multi-center validation and real-time clinical deployment to enhance preventive care.