Comparison of Different Machine Learning Algorithms in the Mental Health Assessment of College Students

Abstract

This paper assesses college students' mental health based on the symptom checklist 90 (SCL-90). In view of the assessment data processing and analysis, the performance of different machine learning algorithms, including random forest (RF), LightGBM3, extreme gradient boosting (XGBoost), in the classification of college students' mental health samples was compared. Moreover, the effect of different hyperparameter optimization methods (grid search, Bayesian optimization, and particle swarm optimization) was compared. The experiment on the SCL-90 assessment dataset found that the optimization effect of grid search was poor, and the highest F1 value and area under the curve (AUC) of the RF algorithm were 0.8914 and 0.9384, respectively, the highest F1 and AUC values of the XGBoost algorithm were 0.9166 and 0.9551, respectively. The LightGBM algorithm optimized by particle swarm optimization showed the best performance in the classification of mental health samples, with an F1 value of 0.9790 and an AUC of 0.9945. It also achieved optimal results when compared to machine learning algorithms such as naive Bayes and the support vector machines. The results prove the reliability and accuracy of the particle swarm optimization-improved Light-GBM algorithm in the analysis of college students' mental health assessment data. The algorithm can be applied in practice to provide an effective tool for the analysis of the mental health assessment data of college students.