A FSA-based approach for mining sequential patterns with user-specified skeletons

Mining time sequential patterns is an important aspect of data mining. The conventional methods of data mining are faced with the disturbance of generating a large number of potential useless rules. An effective method to this problem is that user specifies a group of skeletons for the patterns to be mined. A mining approach based on time window can generate the user-specified patterns by transiting them into finite-state automata. Compared with the general mining methods focusing on sequential data consisting of single event at a point, this paper presents a method of knowledge mining from sequential data consisting of event-sets. Rules mined from database of rice-insect show that this method generates some time sequential patterns about the regularities of rice insect.