Prediction Of Anxiety Disorder in Working Class Students Using Enhanced Machine Learning Algorithms

Abstract

Abstract: anxiety disorder is a mental health condition that affects millions of people worldwide and it make more impact among working-class students due to academic and work stress. Early detection of anxiety is important for easy intervention; the traditional diagnostic methods often fail due to the complex and subjective nature of the symptoms. This study includes a machine learning-based approach to predict anxiety disorder using five classification algorithms: Random Forest (RF), Light Gradient Boosting Machine (LightGBM), Support Vector Machine (SVM), Naïve Bayes (NB), and Decision Tree (DT). To enhance model accuracy and efficiency, Particle Swarm Optimization (PSO) was used for feature selection and hyperparameter tuning. Dataset obtained from Kaggle repository was used, it contains over 140,000 instances with 110 features spanning demographic, academic, and psychological dimensions. Exploratory Data Analysis (EDA) revealed strong associations between anxiety and factors such as work pressure, financial stress, and study satisfaction. PSO identified the most relevant 14 features, improving model interpretability and performance. Among the models evaluated, the PSO-optimized Random Forest achieved the highest accuracy (99%), followed by Decision Tree (97.3%), LightGBM (94.8%), and Naïve Bayes (88.1%). The results confirm that tree-based ensemble methods, particularly when combined with PSO, offer best solutions for anxiety prediction. This research contributes to the field of medical informatics by providing an efficient approach to predict anxiety disorder using AI. The study's findings can help in development of real-time diagnostic tools to aid clinicians and support early interventions in mental health care.