Prediction of Diabetes Risk in Young and Middle-aged Adults: Machine Learning Analysis based on Health Behavior and Physiological Indicators

Abstract

One of the most common chronic illnesses in the world is diabetes. In recent years, the prevalence of diabetes is increasing among young and middle-aged people. This article aimed to making a predict model about diabetes used a balanced data, focusing on health behavior and physiological indicators mainly, from Behavioral Risk Factor Surveillance System (BRFSS) 2015. The data was analyzed by machine learning method and two models have been constructed in this article, which are logistic regression model and random forest model, in order to choose the model with higher accuracy, the accuracy rate, f1-score and a confusion matrix of that two models have been compared. The findings of the study indicated that logistic regression model is better for using in this dataset with higher accuracy. However, its accuracy is 84.42%, which is not high enough for actual use. There are 82 false positives (FPs) and 228 false negatives (FNs) as the prediction outcome of logistic regression model. Based on these findings, it is suggested that more updated variables, different parameters’ selection and other predict models (such as k nearest-neighbor, decision tree etc.) should be considered in model construction.