Interestingness Measures for Association Patterns: A Perspective
Pang‐Ning Tan, Vipin Kumar · University of Minnesota Digital Conservancy (University of Minnesota) · 2000
Association rules are valuable patterns because they oer useful insight into the types of dependencies that exist between attributes of a data set. Due to the completeness nature of algorithms such as Apriori, the number of patterns extracted are often very large. Therefore, there is a need to prune or rank the discovered patterns according to their degree of interestingness. In this paper, we will examine the various interestingness measures proposed in statistics, machine learning and data mining literature. We will compare these measures and investigate how close they reect the statistical notion of correlation. We will show that supportbased pruning, which is often used in association rule discovery, is appropriate because it removes mostly uncorrelated and negatively correlated patterns. Our experimental results veried that many of the intuitive measures (such as Piatetsky-Shapiro's rule-interest, condence, laplace, entropy gain, etc.) are very similar in nature to correlation...