Online Clustering and Outlier Detection

Abstract

Clustering and outlier detection are important data mining areas. Online clustering and outlier detection generally work with continuous data streams generated at a rapid rate and have many practical applications, such as network instruction detection and online fraud detection. This chapter first reviews related background of online clustering and outlier detection. Then, an incremental clustering and outlier detection method for market-basket data is proposed and presented in details. This proposed method consists of two phases: weighted affinity measure clustering (WC clustering) and outlier detection. Specifically, given a data set, the WC clustering phase analyzes the data set and groups data items into clusters. Then, outlier detection phase examines each newly arrived transaction against the item clusters formed in WC clustering phase, and determines whether the new transaction is an outlier. Periodically, the newly collected transactions are analyzed using WC clustering to produce an updated set of clusters, against which transactions arrived afterwards are examined. The process is carried out continuously and incrementally. Finally, the future research trends on online data mining are explored at the end of the chapter.