Application of Diabetes Risk Prediction Using Machine Learning Algorithms

Abstract

Diabetes mellitus is a chronic disease that poses a significant global health burden, requiring effective early detection strategies to reduce complications and mortality. In recent years, machine learning techniques have been widely applied to support medical decision-making, particularly in disease risk prediction. This study aims to compare the performance of several machine learning algorithms for diabetes risk prediction and to implement the best-performing model into a web-based application. The PIMA Indians Diabetes Dataset was used in this study, and data preprocessing was conducted to address class imbalance and improve model performance. Five classification algorithms were evaluated, namely Logistic Regression, Support Vector Machine (SVM), Random Forest, K-Nearest Neighbors (KNN), and Naive Bayes. Model performance was assessed using accuracy, recall, F1-score, and Area Under the Curve (AUC), with a particular emphasis on recall and F1-score due to their importance in medical screening applications. Experimental results show that the SVM model outperformed the other algorithms, achieving higher recall, F1-score, and AUC values. The selected model was then implemented into a web-based application using the Streamlit framework, enabling users to input clinical parameters and obtain real-time diabetes risk predictions. The results indicate that machine learning models, particularly SVM, can effectively support diabetes risk prediction and demonstrate the potential of integrating predictive models into practical healthcare applications.