Identification of pivotal nutritional factors to reduce risk of prediabetes by machine learning

Abstract

Type 2 Diabetes is a major concern in healthcare worldwide with increasingly public health burden. Prediabetes has been considered as a critical stage to reduce the risk of Type 2 Diabetes since its highly correlated with lifestyle changes. The present study explored the risk factor for prediabetes and built a machine learning model to prioritize the important factors for effective prevention. In this study, data was extracted from National Health and Nutrition Examination Survey (NHANES), including over 36,000 people classified to diabetes, prediabetes and healthy. Six machine learning models, namely, random forest, k-Nearest Neighbor (KNN), support vector machine (SVM), gradient boosting, LDA, and Xgboosting were trained for prediabetes risk factor prediction. Random forest eventually outperformed the other four models with an area under curve(AUC) score of 0.71. Also, explorations on the top features that are highly correlated with prediabetes were made. Multiple nutritional factors were identified, including thiamin, folic acid, and caffeine, which shows significant difference between healthy control and prediabetes and the lifestyles regarding these nutrition are recommended to change accordingly.