Detection And Early Warning Of Mental Disorder In Students Based On Time Series Classification

Mental disorder,an obscure psychological disease without obvious symptoms and efficient standards of diagnosis,has long been troubling college students and lead to high probability for all kinds of safety accident.In what way to carry out automatic detection,give early warning and effective intervention of psychological abnormal events is important to campus security and social stability.Therefore,this paper studies the automatic detection of mood disorder and time series classification,tries to solve this problem by only analyzing students’ daily consumption behavior and explore in what pattern they behave,successfully puts forward a mood detection model based Trans-LSTM and developed a mental health early warning system for students using Think PHP.The contribution of this paper can be summarized as follows:(1)The correlation between students’ consumption behavior and mental state has been validated.Considering the problems of low campus data utilization and the difficulty to discover unusual mood,this paper collected consuming records of students in a non-invasive manner,and took the week as time step to count the change track of students’ consumption behavior in a period,such as consumption times,breakfast times,delay breakfast times and so on,then put it into neural network for training.Finally,the experimental result shows that there is a correlation between consumption behavior and mood,and it successfully constructed the abnormal mood detection model based this finding.At the same time,in order to dealing with missing values in dataset,a missing value processing method combining missing information embedding and linear interpolation is designed.The experiment shows that this method has obvious effect in improving classification.(2)As the FCN,a branch of MLSTM-FCN,has limited receptive field and can only run in one direction during feature extraction,an abnormal mood detection algorithm based on Trans-LSTM was designed by replacing it with the multi-head attention,which commonly used in Transformer network and can help improve the ability of feature extraction of MLSTM-FCN.The results indicated that recall of the proposed model is 84%,which is 2%higher than that of the original model.(3)Based on Think PHP framework,a mental illness early warning system is developed to provide customed services,including mental information management,system management and psychological abnormality early warning,for students,teacher and administrator,to maintain campus safety and improve the quality of education.