Prediction of Glycemic Control in Diabetes Mellitus Patients Using Machine Learning

While machine learning has made significant strides in diabetes prediction, glycemic control, a crucial aspect of diabetes management, remains understudied and calls for enhanced forecasting techniques. In addition to economic benefits, proper glycemic control also functions as a preventative measure against potential health complications. This paper aims to provide an accurate and reliable approach to predicting glycemic control in individuals afflicted with diabetes mellitus using advanced machine learning techniques. A vast and comprehensive dataset comprising 77,724 recently diagnosed diabetes patients from Istanbul province of Turkey in the year 2017 has been used in this study. Redundant features were eliminated and class imbalance was mitigated through the implementation of various sampling techniques. A collection of nine machine learning algorithms were utilized in order to predict glycemic control. Among the various trained models, LightGBM and CatBoost demonstrated exceptional performance, outperforming all other models with accuracy and AUC values of 83.36% and 89.49%, and 83.13% and 89.24% respectively. Explainable AI tools, such as LIME and SHAP, were employed to comprehend the predictions of the models and gain insights into important features.