Predicting psychological risk among college students using the Freshman Entrance Psychological Survey: A machine...

With college freshmen under increasing psychological pressures, early detection of those at risk is critical. We applied machine learning to predict their mental health risks.

A psychological screening using UPI and SDS scales was executed on 7211 students admitted in 2023 and 2024 at a university in northern Anhui Province within two months of enrollment. The results showed that 1011 students might have psychological problems. After interviewing and assessing each student individually, 125 students were identified as psychologically unhealthy. LASSO regression is used for feature selection, which is then integrated with logistic regression to create a predictive model. Then constructed a nomogram. Model efficacy was evaluated by receiver operating characteristic curve (ROC), calibration curve, Hosmer-Lemeshow (H-L) goodness-of-fit test, and decision curve analysis (DCA).

Body-mind satisfaction, history of pre-enrollment psychological counseling, history of pre-enrollment psychiatric clinic, Mother died, item 25 ("Have an idea of wanting to die"), UPI score, and SDS score were influencing factors. A model and a nomogram were constructed. The area under the curve (AUC) for the training and validation groups was 0.807 and 0.757, respectively. Brier scores were 0.089 and 0.088 in the training and validation groups, respectively. Calibration slopes of 1.000. H-L tests resulted in χ² = 5.670 (P = 0.684) for training and χ² = 3.842 (P = 0.871) for validation. The DCA showed a high net clinical benefit of the model.

Following preliminary screening, we classified college freshmen and used machine learning to create a predictive model that can forecast psychological risk.