Machine Learning-based Prediction of Depression Risk in College Students: The Role of Diet-Exercise Patterns and Inflammatory Markers

Background: Depression is highly prevalent among college students, with an etiology involving complex interactions between lifestyle factors and physiological mechanisms. Existing studies often examine dietary or exercise factors in isolation, lacking a holistic perspective on behavioral patterns. Furthermore, the independent role of inflammatory markers in depression among young adults remains unclear. To address this, based on data from the National Health and Nutrition Examination Survey (NHANES), this study aimed to identify latent patterns of dietary and exercise behaviors among college students and systematically explore the associations of these behavioral patterns and systemic inflammatory markers with depressive symptoms, as well as their predictive value. Methods: A total of 515 participants aged 18-23 years with an undergraduate education background were selected from four cycles of NHANES (2013-2023). Principal Component Analysis (PCA) and K-means clustering were employed to identify dietary-exercise behavioral patterns. Depression was defined as a Patient Health Questionnaire-9 (PHQ-9) score of ≥10. LASSO regression was used to select predictive features, and six machine learning models-Regularized Logistic Regression, Random Forest, Support Vector Machine (SVM), Elastic Net, Gradient Boosting Machine (GBM), and Naive Bayes-were constructed and their predictive performances compared. Multivariate logistic regression was utilized to analyze the independent effects of behavioral patterns, sleep, and other factors on depression. Results: Two behavioral patterns were identified: the “Sedentary/Low-Intake” pattern (Cluster 1, n=425) and the “High-Energy/Exercise” pattern (Cluster 2, n=90). Univariate analysis indicated that the Neutrophil-to-Lymphocyte Ratio (NLR) and Lymphocyte-to-Monocyte Ratio (LMR) were significantly higher in the depression group (P < 0.05). In the multivariate analysis, female gender (OR = 7.86, P < 0.001) and higher body weight (OR = 1.014, P = 0.023) were independent risk factors for depression, while a sleep duration of ≥9 hours was a protective factor (OR = 0.26, P = 0.017). After adjusting for sleep, the “High-Energy/Exercise” pattern was positively associated with depression risk (OR = 2.53, P = 0.025). Among the machine learning prediction models, the Elastic Net model demonstrated the best performance (Test set AUC = 0.752, Sensitivity = 0.650). Conclusion: Among college students, depressive symptoms are closely associated with gender, body weight, sleep duration, and a specific “High-Energy/Exercise” behavioral pattern, whereas the independent predictive role of inflammatory markers was not significant in the multivariate models. Machine learning models integrating behavioral patterns with demographic characteristics demonstrate robust predictive capability for depression, providing empirical evidence for the development of early screening tools for depression risk on university campuses.