An Efficient Machine Learning Framework for Panic Disorder Detection: Addressing Data Imbalance with SMOTE to Reduce False Negatives

One common mental health condition that can seriously impair daily functioning is panic disorder. It is a type of anxiety disorder characterized by frequent, unplanned panic attacks. Effective treatment requires early and precise detection. With objective, data-driven methods, machine learning (ML) presents a promising way to improve panic disorder detection in early stages and supports broader public health initiatives aimed at mental health awareness and treatment accessibility. This study presents an efficient machine learning framework designed to enhance panic disorder detection by reducing the false negative rate (FNR) through the integration of Synthetic Minority Over-sampling Technique (SMOTE). The proposal employs four ML algoithms including Logistic Regression, Random Forest, K-nearest Neighbor and Decision Tree ensuring that the model is trained effectively on minority class instances using SMOTE. Our results demonstrate a significant reduction in FNR, facilitating earlier detection and timely intervention for patients. We achieved around 71%-92% reduction in the false negative rate, leading to a substantial enhancement in the overall diagnostic performance of the model including precision upto 34%, recall upto 91%, f1-score upto 81% for different ML algorithms. Thus, the importance of this research lies in enhancing diagnostic performance and improving patient outcomes, facilitating a more effective and empathetic approach to mental health diagnosis.