User-centric anomaly detection in activities of daily living

Yahaya, S.W. ORCID: 0000-0002-0394-6112, 2021. User-centric anomaly detection in activities of daily living. PhD, Nottingham Trent University.

Salisu Wada Yahaya 2022.pdf - Published version

Download (11MB) | Preview


The current system for providing care to older adults is not sustainable due to its excessive cost. It places an unbearable financial burden on the government and families and pressure on the workforce due to the demand for human carers. Studies have also shown that older adults prefer to be looked after in their homes rather than in a care facility. An automated system of monitoring can provide much-needed support at a lower cost and give peace of mind to relatives.

The focus of the research reported in this thesis is to investigate the concept of abnormality detection in activities of daily living. More precisely, this work is aimed at proposing a dynamic approach for anomaly detection capable of adapting to changes in human behaviour. Abnormalities in daily activities can be an early indication of health decline. Therefore, early detection can inform the families of the need for intervention. Anomalies are often detected by modelling the existing activity data representing the usual behavioural routine of an individual to serve as a baseline model. Subsequent activities deviating from the baseline are then classified as outliers or anomalies. However, existing approaches suffer from a high rate of false prediction due to the static nature and the inability of the approaches to adapt to the changing human behaviour.

The contributions of the research are reported in four main categories. First, a novel ensemble approach termed "Consensus Novelty Detection Ensemble" is proposed. The outlying activities are predicted by computing their normality score using the internal and external consensus vote and the estimated weights of the models in the ensemble. Activities with a score exceeding a threshold estimated using a statistical method based on data distribution are predicted as outliers and vice versa.

Secondly, a similarity measure approach for identifying the likely sources of the ADL anomalies is proposed. While the models can detect anomalous activities, they are unable to identify the source (cause) of the anomaly. Identifying the anomaly source allows for the development of an adaptive system. The approach is based on a pairwise distance measurement of the features extracted from the activity data. Two approaches for performing the similarity measures are presented, namely, One vs One Similarity Measure (OOSM) and One vs All Similarity Measure (OASM). Features of the data with a higher dissimilarity rate are predicted as the source.

To make the proposed model adaptive to the changes in human behaviour, a novel adaptive approach is proposed based on the concept of forgetting factors. This allows the model to forget (discard) outdated activity data and adapt to the current behavioural patterns by incorporating newly verified data. The data verification can be performed by incorporating human feedback into the system. Two forgetting factor approaches are proposed namely; Forgetting Factor based on Data Ageing (FFDD) and Forgetting Factor based on Data Dissimilarity (FFDA). The data ageing forgetting factor discard old behavioural routine based on the age of the activity data, while in the data dissimilarity approach, this is achieved by measuring the similarity of the activity data.

Lastly, the means of utilising an assistive robot as a communication intermediary is explored for incorporating human feedback into the learning process using hand gestures as a communication modality. Experimental data used for the gesture recognition model is collected using a wearable sensor and a 2D camera. The feasibility of utilising the robotic platform as an exercise coach to encourage physical activity and promote a healthy lifestyle is explored. To this end, an exercise training solution is developed for the robotic platform to coach, motivate and assess the older adults in the recommended physical activities.

Item Type: Thesis
Creators: Yahaya, S.W.
Date: May 2021
Divisions: Schools > School of Science and Technology
Record created by: Linda Sullivan
Date Added: 07 Feb 2022 14:22
Last Modified: 07 Feb 2022 14:22

Actions (login required)

Edit View Edit View


Views per month over past year


Downloads per month over past year