Optimized Neural Networks for Diabetes Classification Using Pima Indians Diabetes Database

Integrating artificial intelligence (AI) into the healthcare sector holds immense potential for transforming the industry, promising notable improvements in diagnostic precision, treatment effectiveness, and overall patient care. This paper explores the detection of diabetes using two types of neural networks - feedforward neural network (FNN) and convolutional neural network (CNN) - with the Pima Indians diabetes database (PIDD). To evaluate the efficiency of the proposed models in diagnosing diabetes, various essential metrics are utilized, including accuracy, precision, recall, F1-score, specificity, receiver operating characteristic - area under the curve (ROC-AUC), log loss, false positive rate (FPR), Youden's index, and Matthews correlation coefficient (MCC). The proposed FNN model achieves an impressive accuracy rate of 82%, outperforming previous methodologies, whereas the CNN displays commendable accuracy of 80.52%. Both models demonstrate superb performance in terms of accuracy, specificity, and AUC, highlighting their effectiveness in binary classification when compared to prior studies. This research provides valuable insights into utilizing advanced machine-learning techniques for the early detection of diabetes.