A MACHINE LEARNING APPROACH TO ASSESS PSYCHOLOGICAL STATE AMONG UNIVERSITY STUDENTS THROUGHOUT THE COVID-19 PANDEMIC: BANGLADESH PERSPECTIVE

The COVID-19 outbreak in Bangladesh had a negative impact on people of all ages. The epidemic's destruction clearly had an effect on people's mental health, especially that of university students. From the beginning of the pandemic in Bangladesh, educational institutes were shut down, complete lockdown condition, unable to get sports and entertainment abilities, which caused the student's psychological health to suffer. Most university-aged students exhibited long-lasting psychological issues corresponding with COVID-19, including significant levels of stress, anxiety, and depression. Predicting the psychological state will indicate a lack of psychological resilience, which will be associated with mental health problems among Bangladeshi university students. Increasing psychological fortitude is essential to ensuring pupils' well-being throughout the epidemic. Through an online survey and several machine learning algorithms, our system predicts the psychological state of Bangladeshi university students. We preprocessed this dataset by cleaning it correctly for the procedure. We utilized hyper-parameter tweaking to extract the features, and then we trained the dataset using a number of classifiers, such as the support vector classifier, random forest, logistic regression, decision tree, naive Bayes, KNN, and gradient boosting. Study suggests that, these algorithms works best in researching on mental health related datasets. Among these several machine learning algorithms, our created dataset of  509 points, comprising support vector classifier (SVC), produced an AUROC score of 0.98, 0.97, and 0.97 for depression, anxiety, and stress states, respectively. Additionally, SVC also delivered respectable outcomes on the open-source dataset we collected for each of the psychological states - depression, anxiety, and stress. Support vector machine (SVM), a supervised machine learning model that employs classification methods, may, in general, produce excellent results when there is a distinct proportion of displacement between classes. By evaluating a dataset we have collected and enhancing the DASS-21 formulation to measure an individual's depression, anxiety, and stress. DAAS-21 is well established screening method for addressing mental health issue. We sincerely considered the ethics and all the data we collected from people are preserved with care and utmost privacy. This study will aid in the growth of research into the area of suicidal thoughts and emotional states.