An Entropy-Based Approach for Anomaly Detection in Activities of Daily Living in the Presence of a Visitor

Abstract

This paper presents anomaly detection in activities of daily living based on entropy measures. It is shown that the proposed approach will identify anomalies when there are visitors representing a multi-occupant environment. Residents often receive visits from family members or health care workers. Therefore, the residents' activity is expected to be different when there is a visitor, which could be considered as an abnormal activity pattern. Identifying anomalies is essential for healthcare management, as this will enable action to avoid prospective problems early and to improve and support residents' ability to live safely and independently in their own homes. Entropy measure analysis is an established method to detect disorder or irregularities in many applications: however, this has rarely been applied in the context of activities of daily living. An experimental evaluation is conducted to detect anomalies obtained from a real home environment. Experimental results are presented to demonstrate the effectiveness of the entropy measures employed in detecting anomalies in the resident's activity and identifying visiting times in the same environment.

Keywords: activities of daily living; activity recognition; anomaly detection; approximate entropy; behavioural patterns; fuzzy entropy; independent living; multiscale-fuzzy entropy; sample entropy.

Figure 1

A schematic diagram of the proposed anomaly detection in activities of daily living in the presence of a visitor.

Figure 2

An illustration of entropy measurement definition.

Figure 3

Different types of entropy measures, presented in chronological order.

Figure 4

Floor plan and sensors location used for data collection in a SmartNTU home environment.

Figure 5

The results obtained by applying Shannon Entropy (ShEn) for anomaly detection in the activities of daily living in the presence of a visitor. The figure also illustrates the threshold value for 65 days.

Figure 6

The results obtained by applying Fuzzy Entropy (FuzzyEn) for anomaly detection in the activities of daily living in the presence of a visitor. The figure also illustrates the threshold value for 65 days.

Figure 7

The results obtained when applying Shannon Entropy (ShEn) for the 9 days with abnormal activity to examine the possible causes of the identified anomalous days based on one-hour time periods. The figure also shows the threshold value for entropy on normal days, which will be used for detecting any hour the anomaly has occurred on anomalous days.

Figure 8

The results obtained from applying Fuzzy Entropy (FuzzyEn) for the 9 days of abnormal activity to examine the possible causes of the identified anomalous days based on one-hour time periods. The figure also shows the selected threshold value for entropy on normal days, which will be used for detecting any hour the anomaly has occurred on anomalous days.

Figure 9

Examples of an identified visitor and irregular sleep using a door sensor with entropy measures for the collected ADL dataset representing: (a) visiting time on day 16, with the time confirmed using the door sensor; (b) irregular sleep on day 49; and, (c) visitor and irregular sleep on day 63.