Using Bi-sets That Characterize Bi-partitions as Features for Classification: an Application to Microarray Data Analysis

As part of the efforts for building a unified Inductive Databases (IDBs) framework, an important step would be to find a way to combine discovered local patterns from the data with global models of predictive nature. In this paper, we investigate the possibility of using bi-sets (local patterns) as features during classification. When searching for bi-sets from Boolean data, despite reasonable frequency constraints, a large number of sets are usually generated. In order to discern which of these bi-sets could be potentially useful as features for classification, we are using a scoring function which includes as parameters the bi-sets coverage and size. After a feature construction process, we perform an experimental evaluation on Huntington’s disesase (HD) microarray data. We apply Predictive Clustering Trees for the problem of distinguishinh between HD and healthy subjects and also for determining the stage of the development of the disease.