Measuring Interestingness – Perspectives on Anomaly Detection

Abstract

We live in a data deluge. Our ability to gather, distribute, and store information has grown immensely over the past two decades. With this overabundance of data, the core knowledge discovery problem is no longer in the gathering of this data, but rather in the retrieving of relevant data efficiently. While the most common approach is to use rule interestingness to filter results of the association rule generation process, study of literature suggests that interestingness is difficult to define quantitatively and is best summarized as, “a record or pattern is interesting if it suggests a change in an established model.” In this paper we elaborate on the term interestingness, and the surrounding taxonomy of interestingness measures, anomalies, novelty and surprisingness. We review and summarize the current state of literature surrounding interestingness and associated approaches.

Keywords: Interestingness, anomaly detection, rare-class mining, Interestingness measures, outliers, surprisingness, novelty