Fighting depression: psychological approaches among Bangladeshi university students

Abstract

In recent years, mental health deterioration cases have increased exponentially indicates that this issue needs our attention. Stressful situations of our daily life such as excessive study load, relationship predicaments, domestic abuse, sexual harassment, and many other reasons cause depression which is often prevalent and ends up causing physical or mental harm. The psychiatrists, psychoanalysts, and counselors are having a tough time dealing with large numbers of cases. They could only help some of the patients as most of them do not have access to them. Moreover, some people cannot bear the cost or feel hesitant to open up to them. In addition, the overall process takes a long time to understand the patient’s condition, and sometimes patients hide information from the counselors that lead to wrong assessment. Sometimes, it is too late to diagnose and treat their depression. As a result, they reach an extremely vulnerable stage and choose the path of self-harm that contributes to the increasing rate of suicide. Analyzing their history and then taking proper measurements can contribute to the treatment of depression. However, the challenge is that human behavior is ambiguous and inconsistent. Therefore, we propose methodologies for diagnosing their mental health conditions by tracking the probable cause of their depression. With the help of deep learning and machine learning, our goal was to analyze large data sets for observing patterns such as age, gender, the causality of depression, the delta of behavior changes, and many other things related to students and excavating things efficiently to help patients. When it comes to making predictions about depression and offering advice, the survey data that we have gathered over the course of this project has been of great assistance. According to the findings of our research, the Random Forest Classier Algorithm is capable of accurately predicting depression with an accuracy of 87%, an f-measure of 86%, and this model is also the best model. In comparison to the other algorithms that we used, such as K-Nearest Neighbor, Support Vector Machine, Gaussian Naive Bayes, Artificial Neural Network, Gradient Boost, and Decision tree, this one performed far better. The recommendation model that was built by us as part of this research project is our novel contribution to this discussion. We will prognosticate to assist the students with mobile application in the near future, so that they feel better with the help of Machine Learning and Deep Learning by inspecting and examining those patterns.