Anomaly Detection Based on Frequent Pattern Mining in Smart Home Devices

As the rapid development of the Internet of Things (IoT) continues, smart homes, which install, connect, and monitor smart devices, are becoming increasingly prominent. However, the short life cycle, improper usage, security vulnerabilities, and potential manufacturing deficiencies of smart devices may lead to the generation of anomalous data that do not accurately reflect real-world conditions, thereby compromising the functionality and reliability of smart home systems. In the literature, several works focus on handling anomalies in smart homes. However, due to the intractably large amount of reported data, detecting abnormal devices and correcting abnormal data remain challenging. In this work, we propose a novel method for detecting device anomalies in smart homes. Specifically, we propose a novel concept of Frequent ACT group, which groups frequently triggered devices together to enhance the efficiency of anomaly detection. Then, we develop a learning model that learns features of frequent ACT groups and a detection algorithm that identifies and corrects abnormal devices and their corresponding abnormal data based on extracted features. Experimental results on five different real-world datasets from the CASAS smart home project illustrate the advantages of our method compared to four efficient baseline methods in the literature in terms of precision, recall, and detection time. Furthermore, we examine the performance of our method under various sets of hyperparameters.