Ensemble Approach: XGBoost, CATBoost, and LightGBM for Diabetes Mellitus Risk Prediction

Diabetes is such a disease that the whole world is troubled. Due to Diabetes Problem, people have to suffer a lot, and they are also losing a lot of money to get rid of it. It isn't easy to detect it before it happens and manage and cure it properly. The reason behind the disease is the lack of Insulin; it is a type of hormone whose defect causes decreases in blood glucose, and the problem of diabetes occurs. Many methods have been used to detect, predict and classify diabetes, including machine learning and deep learning models. Many studies show that diabetes can be prevented if exercise and healthy eating habits are included in the lifestyle. A set of principles was defined and put into practice to detect the diabetes mellitus of the patient’s record in the PID dataset. Many models developed in the earlier study have been compared with the current research in this research work. This development aims to present a model that can diagnose diabetes quickly, thereby saving people both time and money. The widely used dataset is the Pima Indian Diabetes dataset, which is considered for building the model. The dataset require some preprocessing for filling values that are missing and cleaning the outliers. Ensemble learning techniques CATBoost, LightGBM, and XGBoost, are utilized for evaluation. Correct classification rate, sensitivity, and F1 measure were used to analyze the results. LightGBM model gives 96% accuracy and 0.04 Mean Squared Error which is better than state-of-the-art methods.