Machine Learning Techniques for Diagnosis of Panic Disorder in Patients with a Psychiatric History of Anxiety

Laqui Pacaya J.U.; Lucas Benavides J.L.; Pérez Vera Y. · International Journal of Computing (2026) · DOI: 10.47839/ijc.25.3.4840

Resumen

Panic disorder is a significant anxiety-related condition that can severely impact patients' quality of life, especially among individuals with a prior psychiatric history of anxiety. Early and accurate detection is essential for timely intervention but remains challenging due to the multifactorial nature of the disorder. This study aimed to analyze different supervised machine learning approaches to identify the most effective way for diagnosing panic disorder in patients with a history of anxiety. A dataset containing clinical, psychological, and lifestyle variables was analyzed using PyCaret's automated machine learning workflow. Several classification algorithms, including logistic regression, decision trees, random forests, and advanced ensemble models, were trained and evaluated. The results demonstrated that boosting-based models, specifically LightGBM, XGBoost, and Gradient Boosting, achieved the highest performance, with accuracy exceeding 99% and AUC values close to 100%. Interpretability analysis identified key predictive factors such as moderate symptom severity, current stressors, regular exercise, and fear of losing control. These insights provide valuable information for clinical practice and future research, highlighting the potential of interpretable machine learning models as decision-support tools for early detection and personalized intervention. Further validation on extended and demographically varied cohorts is recommended to ensure the models' generalizability and ethical deployment in real-world settings.

Instituciones: Universidad Nacional de San Agustín de Arequipa