Large-margin feature selection for monotonic classification

Monotonic classification plays an important role in the field of decision analysis, where decision values are ordered and the samples with better feature values should not be classified into a worse class. The monotonic classification tasks seem conceptually simple, but difficult to utilize and explain the order structure in practice. In this work, we discuss the issue of feature selection under the monotonicity constraint based on the principle of large margin. By introducing the monotonicity constraint into existing margin based feature selection algorithms, we design two new evaluation algorithms for monotonic classification. The proposed algorithms are tested with some artificial and real data sets, and the experimental results show its effectiveness.