Detecting Anomaly in Smart Homes Based on Mahalanobis Distance

The services provided in smart homes depends heavily on the operation of smart devices which may occasionally have abnormal behaviours due to hardware-failure or improper use. Hence, accurate and quick anomaly detection of devices in smart homes is essential. Due to the explosion of the number of smart devices in smart homes including sensors and actuators, a new anomaly detection method which can deal with a large number of data obtained from smart devices is necessary. Fur-thermore, the line between normal events and abnormal events is really sensitive in some cases, an accurate assessment method is required to reduce the false positive rate in detecting anomaly. In this paper, we propose an anomaly detection method based on Mahalanobis distance to completely solve the above problems. Based on an assumption that complex faults can be detected when actuators are triggered, we group devices which appear together frequently by a key actuator, and then calculate the Mahalanobis distances between states of these frequent groups by KNN models. We also propose a controlling algorithm to judge the failing proportion of devices in order to reduce the false positive rate. The experiment results show that our proposed methods can achieve high detection rates with low false positive rates and small detection time.
^1
1
This research has been done under the research project QG.21.30 “Anomaly detection for IoT devices in smart home environment” of Vietnam National University, Hanoi.