Online Outlier Exploration Over Large Datasets

Traditional outlier detection systems process each individual outlier detection request instantiated with a particular parameter setting one at a time. This is not only prohibitively time-consuming for large datasets, but also tedious for analysts as they explore the data to hone in on the appropriate parameter setting and desired results. In this work, we present the first online outlier exploration platform, called ONION, that enables analysts to effectively explore anomalies even in large datasets. First, ONION features an innovative interactive anomaly exploration model that offers an "outlier-centric panorama" into big datasets along with rich classes of exploration operations. Second, to achieve this model ONION employs an online processing framework composed of a one time offline preprocessing phase followed by an online exploration phase that enables users to interactively explore the data. The preprocessing phase compresses raw big data into a knowledge-rich ONION abstraction that encodes critical interrelationships of outlier candidates so to support subsequent interactive outlier exploration. For the interactive exploration phase, our ONION framework provides several processing strategies that efficiently support the outlier exploration operations. Our user study with real data confirms the effectiveness of ONION in recognizing "true" outliers. Furthermore as demonstrated by our extensive experiments with large datasets, ONION supports all exploration operations within milliseconds response time.