Predicting Anxiety, Depression, and Stress Mental Health Disorders Using Machine Learning Algorithms رسالة ماجستير

Abstract

As Artificial Intelligence rapidly continues to be improved and refined, it has become more possible to help mental health industry owners redefine the meaning of mental health in a more positive way than before. Thus, improving the way to diagnose mental illness in earlier stages, making the treatment more effective. This can be achieved by personalizing the diagnoses and treatment based on the individual’s case characteristics. Mental health is a person's overall psychological well-being, including emotional, psychological, and social factors that affect how they think, feel, and behave. A balance and stability in mental health enable a person to handle normal life stresses, work efficiently, and contribute positively to their community. Since Depression, Stress, and Anxiety are considered the most common disorders, this research aims to employ machine learning algorithms to predict the diagnoses of stress using a dataset collected as part of this work. The dataset consists of around 700 records using an online survey, which was on top of DASS21 (Depression, Anxiety, and Stress international Scale). The data were collected from Palestinian participants, volunteers from the community, and university students. Then data preprocessing has been applied, cleansed from duplicates, and removed useless features like the hand and religion. Also converting textual values into numerical, and due to the imbalance in data, resampling technique has been used to resolve the imbalance to have more accurate results. Five different machine learning algorithms were utilized to analyze the data and achieve the early detection of mental health issues: Random Forest Model, SVM Model, K-Nearest Neighbors Model, XGBoost Model, and Multi-Layer Perceptron (3 Layers) Model. The results for Depression were SVM with the best VI model accuracy at 100%, followed by MLP with 98%, then XGBoost with 95%, Random Forest with 93%, and finally KNN with 77%. The results for Anxiety showed SVM gaining the highest score at 100%, MLP at 96%, both Random Forest and XGBoost at 95%, and KNN at 79%. Also, the results for Stress were 100% for SVM, then Random Forest and XGBoost at 97%, MLP at 96%, and KNN at 78%.