Name
Abstract | ||
|---|---|---|
Anomaly detection of sensor data is a fundamental and active problem, it is involved in many applications especially the Wireless Sensor Network (WSN) where we detect anomaly by the group data. But for anomaly detection of a single sensor, many methods which consider spatial connection of data are not efficient. Point anomaly and pattern anomaly are two types of anomalies. In this work, we analyze the sensor data and find that pattern anomaly plays a dominant role for a single sensor. Moreover, the sudden change of data is found to be the special form of pattern anomaly. As the data collected by a single sensor is ordered by time, we convert the problem to pattern anomaly detection of time series. In this work, a two-phased algorithm OCSVM_KNN based on One Class Support Vector Machines (OCSVM) and K-Nearest Neighbors (KNN) is proposed. And, in the KNN classifier we introduce Complexity-invariant Distance (CID) as the distance measure method between two time-series sequences. Experimental results on real data sets show that our proposed approach outperforms other existing approaches in terms of anomaly detection rate, false alarm rate and misclassification anomaly rate. |
Year | DOI | Venue |
|---|---|---|
2015 | 10.1007/978-3-319-22047-5_18 | Lecture Notes in Computer Science |
Keywords | Field | DocType |
Anomaly detection,Single sensor,Time-series sequence,OCSVM_KNN,One-Class SVM,KNN | Data mining,Anomaly detection,Data set,Computer science,Support vector machine,Active problem,Constant false alarm rate,Classifier (linguistics),Wireless sensor network | Conference |
Volume | ISSN | Citations |
9196 | 0302-9743 | 5 |
PageRank | References | Authors |
0.48 | 11 | 5 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Jing Su | 1 | 5 | 0.48 |
| LONG Ying | 2 | 30 | 9.46 |
| Xiaofeng Qiu | 3 | 23 | 11.41 |
| Shufang Li | 4 | 380 | 26.80 |
| Daowei Liu | 5 | 5 | 0.82 |