Anomaly Detection via oversampling Principal Component Analysis

Abstract Outlier detection is an important issue in data mining and has been studiedin different research areas. It can be used for detecting the small amount of deviateddata. In this article, we use “Leave One Out” procedure to check each individualpoint the “with or without” effect on the variation of principal directions. Based onthis idea, an over-sampling principal component analysis outlier detection methodis proposed for emphasizing the influence of an abnormal instance (or an outlier).Except for identifying the suspicious outliers, we also design an on-line anomalydetection to detect the new arriving anomaly. In addition, we also study the quickupdating of the principal directions for the effective computation and satisfying theon-line detecting demand. Numerical experiments show that our proposed methodis effective in computation time and anomaly detection.