Objectives: Our aim was to overcome the low evaluation accuracy of traditional random sampling methods for college students' mental health, and to use the values of big data of college students' social network behaviors in the prediction and evaluation of their mental health. Methods: We monitored and evaluated college students' mental health through big data analysis. After generating the samples of college students' social network behaviors, a mental health monitoring and evaluation model was established based on a support vector machine (SVM) and decision tree (DT). Then, the DT model was pruned, and input data of the model were optimized by genetic algorithm (GA). Results: The optimal parameter combination was derived for our model. The maximum number of iterations was 60; the smallest number of samples needed for reclassifying internal nodes was 6; the number of samples with the fewest leaf nodes was 30. The mental health scores of most students fell in the interval [0, 6] for unobvious symptoms of mental crisis. The binary classification results of several models were as follows. On anxiety, all models surpassed the accuracy of 60%, except the traditional SVM. The optimal model, ie, Model 5, achieved an accuracy of 86.7%. On depression, all models exceeded the accuracy of 60%, and the GA-optimized DT 5 realized an accuracy as high as 83.1%. On drooping spirit, the optimal model, ie, GA-optimized DT 5, reached an accuracy of 89.5%, which is comparable to that of the GA-optimized SVM 4. Conclusions: The characteristic dimensions extracted by GA are representative. The primary mental states of college students can be estimated quickly and accurately by our model with a low cost of data storage, through the feature analysis of social network behaviors.

No abstract