A novel machine learning method for diet quality evaluation by nutritional ingredients

The current mainstream dietary pattern analysis methods including data-driven and investigator-driven methods have their own limitations.Our goal is to develop a hybrid method to establish the relationship between the intake of various nutrients and the quality of the Chinese diet using machine learning methods. We employed the Chinese Health Eating Index (CHEI) as a apriori expertise to evaluate the diet quality of Chinese people and designed a scheme to predict the CHEI score with nutritional ingredients using machine learning methods. Based on the diet records of 28,000 respondents in the CHNS dataset, seven machine learning regression models were evaluated, and the Gradient Boosting model achieved the best R2 score (0.86). In addition, through separate verification experiments, we found the performance of the proposed method was stable among different population subgroups and the predicted result was reliable.