Accuracy of generalized context patterns in the context based sequential patterns mining
Radosław Z. Ziembiński · Control and Cybernetics · 2011
A context pattern is a frequent subsequence mined from the context database containing set of sequences. This kind of sequential patterns and all elements inside them are described by additional sets of context attributes e.g. continuous ones. The contexts describe circumstances of transactions and sources of se- quential data. These patterns can be mined by an algorithm for the context based sequential pattern mining. However, this can create large sets of patterns because all contexts related to patterns are taken from the database. The goal of the generalization method is to reduce the context pattern set by introducing a more compact and descriptive kind of patterns. This is achieved by finding clus- ters of similar context patterns in the mined set and transforming them to a smaller set of generalized context patterns. This process has to retain as much as possible information from the mined con- text patterns. This paper introduces a definition of the generalized context pattern and the related algorithm. Results from the general- ization may differ as depending on the algorithm design and settings. Hence, generalized patterns may reflect frequent information from the context database differently. Thus, an accuracy measure is also proposed to evaluate the generalized patterns. This measure is used in the experiments presented. The generalized context patterns are compared to patterns mined by the basic sequential patterns mining with prediscretization of context values. Keywords: knowledge discovery, context based sequential pat- tern mining, sequential context pattern clustering, pattern accuracy.