Detecting Abnormal Activity in Daily Living: A Deep Learning Approach with RAT-CNN
Vidhi Jain, Bhagyashri Tushir, Deepak Kumar Sharma, Deepak Gupta · International Journal of Human-Computer Interaction · 2025
Abnormal activity detection in daily living (ADL) is essential for maintaining health and wellbeing, especially for individuals with chronic illnesses or disabilities. We propose a novel method that combines Temporal Convolutional Networks, Residual Attention Blocks, and Markov Distance to capture temporal dependencies and sensor variability for enhanced abnormal activity detection. Experiments on diverse public ADL datasets demonstrate high accuracy and reliability, achieving true positive rates of 94%, 97%, and 91%. The approach offers valuable insights for caregivers and healthcare providers and advances personalized, effective models for real-world applications. Our results highlight the potential of integrating deep learning and statistical techniques to improve quality of life through robust abnormal activity detection.