RegularSearch, a fast performance algorithm for typical testors computation.

In pattern recognition tasks, selecting the appropriate features for a supervised classification problem is a critical task. One theoretical tool used for this purpose is the Test Theory, which identifies feature relevance and reduces the complexity of the training and classification processes. However, searching for irreducible subsets of features, called typical testors, that can differentiate objects belonging to different classes, is an exponential problem in terms of the number of features. To address this problem, various algorithms have been proposed in the literature that use different strategies to avoid unnecessary comparisons. In this paper, is proposed the RegularSearch algorithm to find all typical testors in a dataset associated with a supervised classification problem. The algorithm incrementally searches for candidate subsets of features that are most likely to be associated with a typical testor, reducing the number of comparisons as new features are added to the candidate subset. Our comparisons with the best-performing algorithms reported in the literature demonstrate that RegularSearch is more efficient for processing synthetic and real problem datasets in certain scenarios. [ABSTRACT FROM AUTHOR]