Depression Prediction Among Bangladeshi University Students using machine learning technique

Abstract

This study investigates the application of artificial intelligence techniques to predict mental health breakdowns among university students based on their daily activities. The research uses various data sources, including electronic health records, smartphone usage, and passive sensor data, to create predictive models for detecting depression, suicidal behavior, and other mental health problems. The study found promising results in predicting mental health outcomes with high accuracy, emphasizing feature selection, model optimization, and early intervention using predictive insights. However, gaps in existing research include the need for more comprehensive models, standardization of evaluation metrics, and diverse data sources to capture the complexities of students' experiences. The study aims to address these gaps by creating a predictive model tailored to the university student population, predicting mental health breakdowns based on daily activities. In this work, a survey form was created to collect data on depression, resulting in 1000 responses. The dataset contains 2 parts, 20% is testing data while 80% is training data. The dataset comprises 16 categorical columns, with 10 questions aimed at diagnosing depression independently. Each column represents a different aspect of mental health, lifestyle, or academic experience, without interdependencies between questions. There are 1 dependent and 10 independent or input variables in the dataset. Machine learning methods used such as LR, DT, RF, SVM, Gradient Boosting Algorithm, AdaBoost Algorithm gave 56%, 68%, 68%, 52%, 91% and 89% accuracy respectively. The model is predicted with Gradient Boosting algorithm from the highest accuracy