Margin-closed frequent sequential pattern mining
Dmitriy Fradkin, Fabian Moerchen · 2010
We present a new approach to mining sequential patterns that significantly reduces the number of patterns reported, favoring longer patterns and suppressing shorter patterns with similar frequencies. This is achieved by mining only margin-closed patterns whose support differs by more than some margin from any extension. Our approach extends the efficient BIDE algorithm to enforce the margin constraint. The set of margin-closed patterns can be significantly smaller than a set of just closed patterns while retaining the most important information about the dataset. This is shown by an extensive empirical evaluation on six real life databases.