Classification with Single Constraint Progressive Mining of Sequential Patterns

Abstract

Classification based on sequential pattern data has become an important topic to explore. One of research has been carried was the Classify-By-Sequence, CBS. CBS classified data based on sequential patterns obtained from AprioriLike sequential pattern mining. Sequential patterns obtained were called CSP, Classifiable Sequential Patterns. CSP was used as classifier rules or features for the classification task. CBS used AprioriLike algorithm to search for sequential patterns. However, AprioriLike algorithm took a long time to search for them. Moreover, not all sequential patterns were important for the user. In order to get the right and meaningful features for classification, user uses a constraint in sequential pattern mining. Constraint is also expected to reduce the number of sequential patterns that are short and less meaningful to the user. Therefore, we developed CBS_CLASS* with Single Constraint Progressive Mining of Sequential Patterns or Single Constraint PISA or PISA*. CBS_Class* with PISA* was proven to classify data in faster time since it only processed lesser number of sequential patterns but still conform to user’s need. The experiment result showed that compared to CBS_CLASS, CBS_Class* reduced the classification execution time by 89.8%. Moreover, the accuracy of the classification process can still be maintained.