Fusion in multi-criterion feature ranking

Feature ranking, due to its simplicity and computational efficiency, is a widely used dimensionality reduction technique, especially for large dataset where other methods are computationally too expensive. Conventionally feature ranking is done based on a single ranking criterion. One drawback associated with the conventional, single-criterion ranking is that the ranking order of the features is very much dependent on the ranking criterion used. Such dependence calls for a proper determination of ranking criterion for a given problem at hand However, for most real world problems, the designer often times has no a priori knowledge on what ranking criterion is more favorable, which leads to either a time consuming evaluation process of finding a good criterion or an inferior ranking result due to the use of inappropriate criterion. In this paper, an innovative multicriterion feature ranking (MCFR) scheme is proposed The proposed scheme is based on fusion of a collection of different ranking criteria. The MCFR scheme not only alleviates the difficulty on choosing appropriate ranking criterion, but also improves the reliability of feature ranking results. Implementation details of the proposed feature ranking scheme are provided. Three standard datasets are used for validating the effectiveness of the proposed feature ranking scheme.