Multifunctional Pre-processing Approach for Diabetes Mellitus Prediction Using Ensemble Learning

Diabetes mellitus when untreated can result in a number of health issues. It is a metabolic disease marked by abnormal blood glucose levels. Early detection of diabetes improves a person’s long-term health by halting its progression. Making better medical decisions can be greatly aided through the implementation into practice automated techniques that can precisely identify pre- and diabetic situations. In this paper Synthetic Minority Oversampling Technique and Adaptive Oversampling Technique is used to address class imbalance problem and Chi-square test, Recursive feature elimination methods are involved in detecting and extracting relevant components from unprocessed data. These techniques are applied to the framework and compared using different ensemble learning classifiers such as XGBoost, AdaBoost, LightGBM and CATBoost. The result shows us CATBoost with ADASYN and RFE provides us with better accuracy of 0.96%.