Comparative Study of Machine Learning Pre-processing Techniques for Diabetes Mellitus Prediction

Abstract

Diabetes Mellitus is a chronic and life-threatening disease that requires timely and accurate diagnosis to prevent complications. Machine Learning (ML) techniques have become critical in healthcare for early disease prediction, but their performance heavily depends on the quality of input data. Data pre-processing, including cleaning, feature selection and handling missing values, significantly enhances predictive accuracy. This paper provides a comparative analysis of various pre-processing methods applied to diabetes prediction datasets, including the Pima Indian Diabetes dataset. Techniques such as Average Weighted Objective Distance (AWOD), feature engineering pipelines, wrapper-based feature selection, Support Vector Machines, Random Forests, fuzzy SVM, ANFIS-based imputations, clustering and hybrid approaches are reviewed. Results from previous studies show that optimized feature selection and tailored pre-processing pipelines can boost prediction accuracy to over 98% in some cases. The paper concludes that while no single method is universally applicable, pre-processing plays a crucial role in improving the robustness and accuracy of diabetes prediction models. Future work should focus on adaptive hybrid models combining multiple techniques to handle heterogeneous healthcare datasets.