A Hybrid Convolutional Neural Network Integrated with Federated Learning and Bayesian Optimization (HCNNFLB)Technique for Diabetes Prediction

Diabetes is chronic disease affecting millions worldwide, requiring early diagnosis and continuous monitoring for effective management. Machine Learning (ML) techniques have significantly improved diabetes prediction and classification by analyzing patient data and identifying risk factors with high accuracy. In this study, a Hybrid Convolutional Neural Network integrated with Federated Learning with Bayesian Optimization (HCNNFLB) technique is explored to enhance diabetes diagnosis. The proposed HCNNFLB technique also utilizes feature selection techniques to extract the most relevant biomarker from medical datasets, optimizing prediction performance. Additionally, Federated Learning (FL) is also incorporated to ensure privacy-persevering training across distributed healthcare institutions without sharing sensitive patient information. The proposed HCNNFLB technique outperforms conventional ML models by improving accuracy 94\%, precision 93\%, recall 91%, F1 score 92%, AUCROC 0.95, training time of $120 s$.