Title:Finding Maximal Non-Redundant Association Rules in Tennis Data

Abstract:The concept of association rules is well--known in data mining. But often redundancy and subsumption are not considered, and standard approaches produce thousands or even millions of resulting association rules. Without further information or post--mining approaches, this huge number of rules is typically useless for the domain specialist -- which is an instance of the infamous pattern explosion problem. In this work, we present a new definition of redundancy and subsumption based on the confidence and the support of the rules and propose post-- mining to prune a set of association rules. In a case study, we apply our method to association rules mined from spatio--temporal data. The data represent the trajectories of the ball in tennis matches -- more precisely, the points/times the tennis ball hits the ground. The goal is to analyze the strategies of the players and to try to improve their performance by looking at the resulting association rules. The proposed approach is general, and can also be applied to other spatio--temporal data with a similar structure.