Differentially Sampled Anomaly Detection System based on Outlier Identification Technique

Anomaly detection is the process of identifying unusual behavior and also a small group of instances that deviate remarkably from the existing data. The real world application of anomaly detection includes intrusion or credit card fraud detection that requires a most efficient framework for identifying the deviated data instances. The technique called Principal Component Analysis (PCA) which require large amount of computation memory requirements and therefore it is not suitable for large scale data like online applications. Therefore a new technique called online Oversampling Principal Component Analysis (osPCA) algorithm along with online updating technique is used for detecting the existence of outliers from a large number of data. When oversampling a data instance the online updating technique enables the osPCA to update the outlier identification effectively without solving the eigenvalue decomposition. The feasibility of osPCA provides more efficient and accurate results. The work extends by detecting outliers from high dimensional dataset using some clustering techniques with lesser time consumption.