A Study of Factors Associated with Nutritional Status of Loei Rajabhat University Students by Machine Learning

Abstract

     Objective: This study aimed to identify factors associated with the nutritional status of students at Rajabhat Loei University and compare the effectiveness of machine learning algorithms in predicting nutritional status.

     Methods: This cross-sectional analytical study involved a sample of 1,020 participants categorized into three nutritional status groups using Body Mass Index (BMI) criteria: undernutrition, normal nutrition, and overnutrition/obesity. Ordinal logistic regression was used to analyze the relationships between independent variables and nutritional status. The prediction of nutritional status using machine learning was conducted using three datasets: training, validation, and testing sets. The algorithms evaluated included Support Vector Machine (SVM), Naïve Bayes (NB), Decision Tree (DT), Random Forest (RF), CART, C5.0, Gradient Boosting Machine (GBM), Logistic Regression (LR), and k-nearest neighbors (KNN).

     Results: The research findings revealed that 13.53% of participants had undernutrition, 60.78% had normal nutrition, and 25.69% had overnutrition or obesity. Factors significantly associated with nutritional status included gender, year of study, perception of weight change, nutritional knowledge, and self-management of nutrition. The effectiveness of nutritional status prediction using machine learning showed that the C5.0 algorithm had the highest efficiency, with a correct prediction of 60.78 percent (S.D.=0.00). However, when considering the efficiency of classification by group using the confusion matrix, the C5.0 algorithm fails to classify the sample class 1 and 3. Therefore, the efficiency of the Logistic Regression algorithm was considered, which had the second highest efficiency, with a correct prediction of 59.51 percent (S.D.=3.03), slightly different from the C5.0. It had the highest efficiency in classifying the groups with the reliability, recall, and counterbalance of 0.31, 0.33 and 0.27 respectively. Therefore, in this research, the Logistic Regression algorithm was considered the most appropriate machine learning method for further development. This study lead to a more effective classification of nutritional status and prevention and correction of nutritional problems.