Comparative Assessment of Four Machine Learning Algorithms for Diabetes Risk Prediction

Abstract

The combination of machine learning algorithms with the healthcare industry has become increasingly close, significantly advancing disease prediction capabilities. Each machine learning model has its own fields and scenarios. This study focuses on the predictive effects of four machine learning models on diabetes. They are Random Forest (RF), CatBoost, XGBoost, and LightGBM. The training set and test set of this study are from CDC's 2015 Behavioral Risk Factor Surveillance System. Python is used for the processing of data sets and the visual presentation of results. The final result uses the five aspects of precision, recall, accuracy, Fl-score, and AUC to quantitatively evaluate the prediction effect of different models. The results show that the comprehensive prediction effects of CatBoost, XGBoost, and LightGBM are similar and higher than RF. Furthermore, LightGBM with the highest recall rate is used to analyze the importance of feature variables. The purpose of this study is to provide a reference for the selection of ensemble learning models for predicting chronic diseases such as diabetes. Not only can this improve the prediction efficiency of chronic diseases, but also help early screening of high-risk groups.