Anxiety and Depression Mental Health Factors: Predicting At-Risk Individuals using Supervised Learning Techniques

Abstract

Mental health is a crucial component of human well-being, yet conditions such as anxiety and depression continue to affect individuals globally, often remaining undetected due to stigma and limited access to care. This study applies supervised machine learning techniques to predict individuals at risk of anxiety and depression using the Anxiety and Depression Mental Health Factors dataset sourced from Kaggle. The dataset consists of 1,200 survey responses encompassing demographic, lifestyle, medical, and psychosocial variables. Data preprocessing involved normalization, factor conversion, and the creation of binary “High” and “Low” risk labels based on a threshold score of 12, aligned with established clinical measures such as PHQ-9 and GAD-7. Three supervised learning models—Random Forest, Decision Tree, and Naive Bayes—were developed using an 80:20 train-test split and 5-fold cross-validation. Among these, the Random Forest model performed best, particularly in predicting depression (F1 = 0.473, Recall = 0.452, AUC = 0.545). Anxiety prediction, however, exhibited weaker performance across all models, indicating potential limitations in feature diversity or data balance. Feature importance analysis identified stress level, sleep hours, financial stress, and social support as significant predictors of mental health risk. The findings suggest that while machine learning models show promise in supporting early mental health detection, their accuracy depends heavily on data quality and feature representation. Future work may incorporate more comprehensive datasets and hybrid model approaches to strengthen predictive performance and support data-driven mental health interventions.