Anomaly detection method using center offset measurement based on leverage principle

Anomaly detection is an important branch of data mining and has been researched within diverse research areas and application domains. Many existing unsupervised anomaly detection methods have high computational complexity and more adjustable parameters. In addition, the proportion of anomalies needs to be estimated by long-term experience in most methods to set the threshold that distinguishes normal instances and anomalies, which makes the algorithms more subjective. This paper presents an anomaly detection algorithm based on the leverage principle. When detecting a testing instance, we copy it in large quantities and take these replicated data into the training dataset. The anomaly degree of the testing instance can be assessed by measuring the offset of the dataset center. Meanwhile, an adaptive threshold setting method using the golden ratio is proposed to solve the subjectivity in distinguishing normal instances and anomalies. In the experiments, we compare the anomaly detection algorithm proposed in this paper with other eight detection methods and report the experimental results in terms of AUC, the F1 of the anomaly classes and the running time. The results show that our algorithm can achieve high detection performance with high efficiency, and the proposed threshold setting method also has strong practicality in unsupervised anomaly detection.