The Hybrid Deep Learning ANN-CNN Model for Enhancing Diabetes Prediction

Abstrak

Diabetes mellitus is a global chronic metabolic disease that poses a serious threat to human health. Accurate and early prediction of diabetes is essential for effective medical treatment and long-term disease management. In this study, we propose a deep learning–based framework as a novel approach for diabetes prediction using a large-scale dataset containing more than 6,000 patient records. Several deep learning architectures, including Artificial Neural Networks (ANN) and Convolutional Neural Networks (CNN), are examined to determine the most effective model for prediction tasks. Building on the strengths of both methods, this research introduces a hybrid ANN–CNN architecture designed to leverage ANN’s capability in learning nonlinear relationships and CNN’s efficiency in extracting high-level feature patterns. Extensive data preprocessing and feature extraction were conducted to enhance model performance and ensure reliable outcomes. Experimental results demonstrate that the hybrid ANN–CNN model achieved the highest prediction accuracy of 91.4%, surpassing standalone ANN (86.2%) and CNN (88.9%) models. These findings highlight the potential of hybrid deep learning frameworks in improving clinical decision support systems, enabling more accurate risk assessment and early intervention for diabetes. The results further indicate that integrating complementary neural network structures can significantly enhance predictive performance in complex medical datasets.