Mining Interesting and Contiguous Maximal Sequential Patterns on High Dimensional Sequences

Previous methods have presented convincing arguments that mining complete set of patterns is huge for effective usage. A compact but high quality set of patterns, such as closed patterns and maximal patterns is needed. Most of the previously maximal sequential pattern mining algorithms on high dimensional sequence, such as biological data set, work under the same support. In this paper, an efficient algorithm MM-PrefixSpan (Maximal and Multi-Support-based PrefixSpan) for mining maximal patterns based on multi-support is proposed. Thorough performances on Beta-globin gene sequences have demonstrated that MM-PrefixSpan consumes less memory usage and runtime than PrefixSpan. It generates compressed results and two kinds of interesting patterns.