Study of First Local Maximum of Confidence in Mining Sequential Patterns

Abstract

Sequential data mining is increasingly important in many domains. WinMiner is a constraint-based algorithm to retrieve frequent episodes and association rules of high confidence and to search first local maximum (FLM) - rules. An algorithm for mining FLM rules from sequential dataset is implemented and is applied to several datasets of different origins. The experiments show that FLM rules are rare in randomly generated dataset and loosening the mining constraints leads to the increase of numbers of FLM rules. Correlations or dependencies among the constituent events introduced into the randomly generated dataset can dramatically increase numbers of FLM rules.