An anomaly detection method based on feature mining for wireless sensor networks. (13th July 2021)

• Record Type: Journal Article
• Authors: Ding, Xuefeng; Feng, Wen
• Is Part Of: International journal of sensor networks. Volume 36:Number 3(2021)
• Page Start: 167
• Page End: 173
• Publication Date: 2021-07-13
• Subjects: WSNs -- wireless sensor networks -- abnormal detection -- abnormal data -- time series -- feature mining -- dimensionality reduction -- confidence interval; Sensor networks -- Periodicals; 681.2
• Languages: English
• ISSNs: 1748-1279

Abstract

To overcome the problems of large errors in data feature acquisition and long detection delays in traditional detection methods, this paper proposes an anomaly detection method based on feature mining for wireless sensor networks (WSNs). In our method, dimensionality reduction is performed on the data, all wireless sensor nodes are classified by a hybrid immune method, and data features are mined through vector set recognition. Moreover, the confidence interval is set by a time series, and the effective detection of abnormal data is conducted by comparison. The experimental results show that the maximum error of anomaly data collection is only 1.9%, the maximum time cost of anomaly detection is 8.4 s, and the P-R value is high, indicating that the proposed method is effective.