An Anomaly Detection Scheme with K-means aided Extended Isolation Forest in RSS-based Wireless Positioning System

Over the past years, tremendous progresses have been achieved for wireless positioning system based on Received Signal Strength (RSS). However, most researches assume ideal RSS data, which ignore the anomalies that generally exist due to the interference in signal and instrument malfunction. To reduce the negative influence of anomalies on system performance, anomaly detection is regarded as an important preprocessing technique. To better distinguish anomalies from the RSS data that is distributed in multiple blobs, this paper proposes a two-step anomaly detection scheme called K-means aided Extended Isolation Forest (KEIF). The first step is to exploit K-means to cluster the received data according to the RSS features. Then, based on the positions of source node, Extended Isolation Forest (EIF) is employed for each cluster to obtain anomaly scores, which represent the isolated degree of data points. The data with scores higher than a threshold is considered as anomalies. We verify our proposed scheme in a RSS-based fingerprinting wireless positioning system, and the experiments demonstrate that the real dataset processed by KEIF can effectively improve the positioning accuracy, compared with the original dataset without anomaly detection and the datasets processed by other existing anomaly detection schemes.