LAUSR

Objective: To establish vitamin D classification models for Chinese elderly using machine learning techniques. Methods: Based on the datasets of 2010-2012 Chinese nutrition and health surveillance, the basic information and physical exercise of the subjects were collected. The dietary intake of the subjects was collected by using 3 days-24 hours dietary review method and food frequency method. The normal and insufficient vitamin D was outcome variables. Several machine learning techniques, such as random forest, kernel support vector machine, extreme gradient boosting, and ensemble learning were used to establish vitamin D classification models. Results: Based on the two groups of dietary survey data obtained by using 3 days-24 hours dietary review method and food frequency method, the accuracy of vitamin D classification models for Chinese older people were 0.71 and 0.62, with F1 about 0.82 and 0.73, respectively. The area under curve was 0.58 and 0.57 after adjusting parameters and applying ensemble learning method. Age, gender, intake of vegetables, aquatic product and grains, daily housework, and exercise were important factors to the classification of vitamin D among Chinese elderly. Conclusion: Machine learning techniques could be used to establish vitamin D classification models for Chinses elderly, of which random forest and ensemble learning could be more suitable for the construction of vitamin D classification models.