Tracking feature relevance with the plaid model in continuously changing datasets

Selecting relevant features in data modeling is critical to ensure effective and accurate prediction of future effects. The problem becomes compounded when the relevance of previously selected features cannot be guaranteed due to changes in the underlying dataset. We propose an algorithm based on the statistical plaid model for the discovery and tracking of feature relevance scores in datasets that undergo changes with time. Initially, the algorithm discovers a set of biclusters based on plaid assumptions, and then selects sets of features that define individual biclusters as candidates whose relevance scores are computed and tracked over time. The proposed algorithm was tested on a real changing gene expression dataset, and the results indicate that it was able to successfully identify and track sets of feature relevance scores in the underlying changing dataset.