Enhancing Panic Attack Predictions: Addressing Data Imbalance in Machine Learning Models

Panic disorder (PD) is a prevalent mental health condition characterized by recurring panic attacks, affecting millions globally. Despite advancements in machine learning (ML) for mental health predictions, data imbalance remains a significant challenge, often impairing model performance. This study aims to enhance the accuracy and reliability of panic attack prediction models by implementing a novel data balancing strategy. Using the College Student Mental Health Dataset, ML models—Decision Tree and Random Forest—were evaluated before and after applying data balancing techniques. Initial findings revealed suboptimal model performance due to imbalanced data, with AUC values of 0.434 (Decision Tree) and 0.527 (Random Forest). Post-balancing, significant improvements were observed, with AUC increasing to 0.768 and 0.612, respectively. This study highlights the critical role of data balancing in optimizing ML models, even with limited datasets, and provides a foundation for future research on scalable predictive technologies for mental health disorders. The findings underscore the potential of simple yet effective strategies in advancing early detection systems for mental health conditions.