Classification of Early Onset of Diabetes with Machine Learning, Deep Learning Algorithms, and Bayesian Networks

Diabetes is a chronic disease that cannot be permanently cured, making early detection critical. Our research utilized Data Mining, Machine Learning (ML), Deep Learning, and Bayesian Networks to predict diabetes. We used the Pima Indian Diabetes Dataset (PIDD), which was obtained from the UCI ML Repository and contains data from 768 patients across nine distinct variables. Seven ML algorithms were applied to this dataset to predict diabetes, and we found that Random Forest (RF), Support Vector Machine (SVM), and AdaBoost (AB) performed well. Additionally, we explored BN to model probabilistic relationships between diabetes risk factors, which further contributed to prediction accuracy. We also developed Neural Networks (NN) with varying hidden layers and epochs, discovering that an NN with three hidden layers achieved an accuracy of 89.90%.