Feature Selection with Multi-Cost Constraint

Cost-sensitive learning extends classical machine learning and data mining by considering various types of costs, of the data. Due to money limited, we also have a constraint on the cost for selecting feature and tradeoff between the test costs and misclassification costs. However, the precious works seldom involve the tradeoff problem under the multi-cost constraint. In this paper, we introduce the misclassification costs into the cost constraint problem firstly, and propose a quadratic heuristic algorithm to deal with the minimal feature selection with multi-cost constraint problem. The goal is to obtain a feature subset with minimal average total cost, which includes test costs and misclassification costs. Experimental results indicate the proposed algorithm is effective and efficient.