Various Sequence Classification Mechanisms for Knowledge Discovery

Abstract

Sequence classification is an efficient task in data mining. The knowledge obtained from training stage can be used for sequence classification that assigns class labels to the new sequences. Relevant patterns can be found by using sequential pattern mining in which the values are represented in sequential manner. Classification process has explicit features but these features are not found in sequences. Feature selection techniques are sophisticated, but the potential features dimensionality may be very high. It is hard to find the sequential nature of feature. Sequence classification is a more challenging task than feature vector classification. Sequence classification problem can be solved by rules that consist of interesting patterns. These patterns are found in datasets that have labeled sequences along with class labels. The cohesion and support of the pattern are used to define interestingness of a pattern. In a given class of sequences, interestingness of a pattern can be measured by combining these two factors. Confident classification rules can be generated by using the discovered patterns. Two different approaches to build a classifier are used. The first classifier consists of an advanced form of classification method that depends on association rule. In the second classifier, the value belonging to the new data object is first measured then the rules are ranked.