Classification of Student’s Mental Health using Bidirectional Long Short-Term Memory with Self Attention Mechanism

In this era, mental health issues among university students are a growing concern as it affects the academic performance, personal well-being, and societal stability. Even though previous researchers suggested various tradition methods for classification of student’s mental health still there are issues due to the sequential dependencies in students' behavioral patterns and lack of ability to capture temporal trends in mental health indicators. Therefore, this research proposes a robust model namely Bidirectional Long Short-Term Memory with Self Attention Mechanism (Bi-LSTM-SA). Initially, the data is collected from a cross-sectional survey then fed into the preprocessing stage which employs Multivariate Imputation by Chained Equations (MICE) and Synthetic Minority Over-Sampling Technique (SMOTE) for incorporated for handling missing values and data imbalance issue respectively. Further, a hybrid feature selection approach is employed which includes Boruta Algorithm and step wise selection for optimal selection of features. Then, the proposed Bi-LSTM-SA is imposed to capture both past and future context in student health behavior patterns and classify students into mental health risk categories. The proposed Bi-LSTM-SA acquired greater accuracy with 0.978 when compared to the existing Deep Neural Network (DNN) respectively.