Абстрактный

Online Updating osPCA Technique for Outlier Detection

J. Shankar Babu, Y. Ramya Sree

Anomaly or outlier detection plays an important part in detecting intrusions in real world applications such as credit card frauds, customer behavior changes, defects in manufactured goods or devices etc., and to find out the deviated data instances. In this paper, to overcome the problems in anomaly detection we propose an algorithm of online oversampling principal component analysis osPCA. By using online updating technique, we detect the existence of anomalies in large scale data. Our approach is efficient and more interested in large scale or online problems .We can extract the principal direction of data by oversampling the target instance. Since the anomaly of the target instance is determined according to the difference of resultant principal eigenvector, the osPCA need not to perform eigen analysis. Our proposed construction is more special for online applications. When compared with other anomaly detection algorithms and PCA methods, our tentative outcomes confirm the achievability of our anticipated method is both efficient and accurate.