Research On Prediction Of College Students’ Mental Health Based On CatBoost Algorithm

Objective:Based on the mental health-related data collected from the self-acceptance and growth workshop,this study intends to use the Cat Boost algorithm to construct a prediction model of college students’ mental health status,to achieve the prediction of mental health status through the Cat Boost model,to identify individuals at high risk of psychological problems,and to study the role of the self-acceptance and growth workshop in college students’ mental health,so as to provide a theoretical basis for This study will provide new ideas for large-scale screening of psychological problems in colleges and universities and even in society,and provide theoretical basis for more rational and targeted mental health education and intervention measures.Methods:From November to December 2020,a structured questionnaire was administered to all new master’s students participating in the Self-Acceptance and Growth Workshop using the Symptom Checklist 90,Self-Acceptance Questionnaire,Chinese Big Five Personality Inventory brief version,Psychological Stress Scale for College Students,Purpose-in-Life,Meaning in Life Questionnaire,Parental Bonding Instrument,and Pittsburgh Sleep Quality Index.The data set was divided into a training set and a test set in the ratio of 7:3.The collected variables were screened for optimal features as input variables using recursive random forest,and whether the mental health status was positive or not as output variables.The mental health prediction models were constructed in the training set by machine learning algorithms Cat Boost,support vector machine and logistic regression algorithms,and in the test set using accuracy,accuracy,recall,F1 value and AUC value for model performance evaluation and comparison;ranking and attributing the importance of each feature using SHAP based on Cat Boost model results;comparing the changes in mental health status of the group before and after the workshop intervention.Results:Among 513 people,180 cases were screened positive and 333 cases were nonpositive,with a positive rate of 35.1%;37 variables were collected,and through screening,a total of 20 variables were finally entered into the model training,namely: emotional stress,self-acceptance,life feelings,neuroticism,sleep quality,career selection stress,selfevaluation,interpersonal stress,relationship stress,academic stress,school environment stress,meaning seeking The results of Cat Boost in the test set were: accuracy 85.71%,precision 80.77%,recall 77.78%,F1 value 79.25%,AUC value 0.9148;the results of support vector machine in the test set were: accuracy 83.11%,79.17% accuracy,70.37%recall,74.51% F1 value,and 0.9067 AUC value;the results of logistic regression in the test set were: 82.47% accuracy,64.96% accuracy,70.37% recall,73.79% F1 value,and0.9056 AUC value,which shows that the Cat Boost model has better performance in predicting the mental health status of college students.Emotional stress,self-acceptance and life feelings had higher SHAP values;the higher the values of emotional stress and neuroticism,the higher the risk of having positive mental health,and the higher the values of self-acceptance and life feelings,the lower the risk of having positive mental health.The SCL-90 total score,obsessive-compulsive symptoms,interpersonal sensitivity,depression,anxiety,other,sleep quality,self-acceptance and self-evaluation factors were statistically significant before and after participation in the workshop(t=4.244～-4.896,P<0.05),while the other factors were not statistically significant(P>0.05).Conclusions:The performance of the college student mental health prediction model based on Cat Boost model is good,and the model can be incorporated into the scale screening decision of college students’ mental health status to provide new ideas for early identification of individuals with abnormal mental health;the visual interpretation analysis of SHAP shows the importance and its role of features such as emotional stress,selfacceptance and life feelings in the prediction model of mental health status,which is very important for The visual interpretation analysis of SHAP shows the importance of the characteristics of emotional stress,self-acceptance and life feelings in predicting the model of mental health status and their roles,which provides more targeted opinions for conducting regular mental health education and subsequent psychological interventions,and is very important for improving the mental health of college students;the selfacceptance and growth workshop has a positive effect on college students’ self-acceptance and mental health,and is worth further promotion.