A Model for Detecting Abnormality in Activities of Daily Living Sequences Using Inverse Reinforcement Learning

ADL abnormality detection has been the focus of many recent healthcare studies, some of which addressed the issue by using deep learning techniques. In this paper, we provide a novel approach for examining ADL sequences to detect meaningful deviations from the individual's routine behavior. This approach can benefit older adults in several ways, including timely care, early detection of health conditions to stop them from getting worse, reducing the burden of monitoring on family members, and maximizing self-sufficiency without interfering with daily activities. We present an Inverse Reinforcement Learning (IRL)-based method for detecting behavior abnormalities in older adults through the analysis of ADL sequences. To do this, we model the problem of abnormality detection in behavior sequences as a Higher-order Markov Chain model. Using the IRL method, from observed trajectories of behavior, we infer the reward function that drives the individual to perform ADLs. The inferred reward function will then be utilized to detect potential behavior abnormalities through a threshold-based mechanism.