Comparative Analysis of Logistic Regression, Decision Trees, and Neural Networks for Diabetes Prediction

Diabetes is a chronic metabolic disease that affects millions of people worldwide and can cause complications including kidney failure, heart disease, and vision loss. Recognizing and predicting the condition early is crucial for effective management and avoiding these complications. The objective of this study is to evaluate the performance of three different machine learning models (Logistic Regression, Decision Trees, and Neural Networks) in predicting diabetes risk using a dataset that contains health indicators such as age, BMI, glucose, and insulin levels. We trained and evaluated each model using accuracy, recall, F1-score, and AUC-ROC metrics, hyperparameters tuning, and feature importance analysis using SHAP values. Neural Networks achieved the highest accuracy of 84%, followed by the other models, Decision Trees and Logistic Regression (80% and 76% respectively). Age, BMI, and fasting glucose, among other key health indicators, were identified as significant predictors in each of our models. Therefore, they imply that despite the predictive accuracy of the NNs, it is due to the decision transparency that better suits the clinical context by Decision Trees. These insights will assist health care professionals in deciding optimal models for early detec65tion of diabetes and personalized care strategies.