Highly unbalanced data sets occur frequently in many practical applications and quite often the class of interest in such data sets is just a minority class. Like most standard machine learning methods, traditional rough sets based rule learning algorithms do not usually work well on highly unbalanced data sets. In this paper, we present a minority class rule learning algorithm for a highly unbalanced inconsistent data set where the class of interest is the minority one. The proposed algorithm pivots on discovery of the main features that discriminate the minority class from majority classes by finding the so called dominant minority subset. An illustrative example and a real application to customer churning prediction in Telecom are given to show the effectiveness of the proposed algorithm.
展开▼
