Enhancing Web-Based Diabetes Prediction Using Random Forest Optimization and SMOTE

Abstrak

This research develops a website-based diabetes early detection information system using random forest method and synthetic minority oversampling (SMOTE) technique. This system is designed to predict the risk of diabetes based on user-inputted symptoms. Flask framework is used to optimize web application development. System testing was conducted using black box testing method and validation by medical experts, showing accurate prediction results. The use of the Flask framework facilitates the integration of modeling and user interface development. The data used in this research was balanced using SMOTE, resulting in a prediction accuracy of 96%. The results show that this information system is effective in providing early prediction of diabetes risk and can be a tool for the community to increase awareness of the importance of periodic health checks. The system also provides information that can be used to take preventive measures against diabetes mellitus, supporting government efforts to improve public health.