Comparison of Deep Learning Algorithms Performance In Diabetes Prediction

Diabetes mellitus is a chronic metabolic condition marked by sustained elevations in blood glucose levels, which, without proper management, can progress to serious and wide-ranging complications. Early diagnosis and prediction of diabetes are essential for effective treatment and control. This study seeks to assess and compare the performance of various deep learning algorithms in predicting diabetes using the Pima Indians Diabetes Dataset (PIDD). The deep learning models utilized include Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), Artificial Neural Networks (ANN), and Deep Neural Networks (DNN). Data preprocessing techniques, such as addressing missing values, outliers, and imbalanced datasets, are applied to enhance model performance. The evaluation metrics employed include accuracy, recall, specificity, precision, ROC-AUC and F1-score. The findings reveal that CNN outperforms the other models, achieving the highest accuracy (82.70%), precision (83.33%), recall (85.00%), specificity (80.00%), and F1-score (84.16%), with an AUC of 0.92 . In contrast, DNN, ANN, and LSTM show lower performance, with CNN being the most effective model for diabetes prediction. This research highlights the promising role of deep learning approaches in diabetes prediction and indicates that future investigations could focus on developing hybrid architectures and implementing additional optimization strategies to further improve predictive performance.