Towards the Development of an Adaptive System for Detecting Anomaly in Human Activities

As the global population of older adults increases, measures are put in place to monitor their well-being, promote independent living and improve their quality of life. Among those measures are in-house monitoring system that allows for the collection of data in a non-invasive form to identify the activities of daily living of the older adults and detect abnormalities in their daily routines. Abnormalities can be an early sign of health decline or other related challenges, thereby informing the family and carers of the need for intervention. However, existing anomaly detection systems are unable to adapt to the dynamic nature of human activities which are subject to changes due to different factors, resulting in an increased false prediction rate. To address this deficiency, the anomaly detection system must be adaptive to the changes in human routines as well as factors leading to the changes. This paper presents a consolidation of the achievements recorded in the development of an adaptive anomaly detection system. This system consists of a data collection and interpretation component, anomaly detection component and a feedback component. An ensemble of novelty detection model based on a consensus approach is utilised for the anomaly detection while the feedback component is based on a gesture recognition model implemented on an assistive robot platform. The results of our proposed approach for anomaly detection and gesture recognition performs better when compared to other existing approaches. The obtained results for the different system components show the potential of the system for in-house monitoring of older adults.