AI-Driven Panic Detection system using efficient learning model

Abstract Wearable devices such as smartwatches are becoming useful tools for monitoring different aspects of human health. In this study, an AI-based approach is developed to identify panic events using physiological and activity-related data obtained from commercial smartwatches. The study uses six features: gender, age, weight, heart rate, step count, and physical activity type. A dataset of 16,200 observations collected from 27 participants was used, with separate training and testing sets already provided. Before training the models, duplicate records were removed, categorical data was converted into numerical form, and MinMax normalization was applied. Different machine learning algorithms, along with three deep learning models, were tested and compared using Accuracy, Precision , Recall, F1-score, ROC-AUC, Matthews Correlation Coefficient (MCC), and Cohen’s Kappa. The results revealed that machine learning models performed better than deep learning models for the selected features. K-Nearest Neighbours (KNN) achieved the best accuracy of 96.05%, while Random Forest and Decision 1 Tree also produced good results. This work shows that basic information available through smartwatches can be explored for panic event detection and may help in developing wearable-based systems for continuous mental health monitoring.