Multi-criterion Strategy Of Attribute Reduction And Its Effectiveness

As an effective intelligent information processing method,rough set theory is mainly used in knowledge discovery,decision analysis,data mining and other fields.With further research,researchers put forward many rough set extension models.Among them,neighborhood rough set model and fuzzy rough set model have been widely used.Neighborhood rough set model has a great advantage in dealing with the uncertainty problems;Fuzzy rough set model makes the rough set theory have a better ability in dealing with the fuzzy problems.No matter what kind of model,attribute reduction is still an important problems in the field of rough set,which aims to eliminate redundant conditional attributes without affecting the classification performance.However,the criterion of computing reduct is frequently determined by one single measurement,which is not only difficult to obtain satisfactory stability,but also the selected attribute subset may not the best attribute subset.To solve these problems,this paper will carry out the research work from the perspective of multi-criterion in attribute reduction.The main research contents are as follows:(1)In order to reduce the attribute of the neighborhood information system with the coexistence of symbolic and numerical attributes,the neighborhood entropy in neighborhood information system is defined from the perspective of information theory.Through considering the influence of knowledge uncertainty and set uncertainty on the measurement of attribute importance.Combining the neighborhood conditional entropy with approximation quality,a new attribute importance measurement is defined and used as the criterion of attribute reduction.The experimental results show that the new reducts can not only effectively reduce the neighborhood conditional entropy,but also effectively improve the accuracy of neighborhood classifier.(2)With the help of neighborhood rough set model,an ensemble attribute reduction strategy of neighborhood decision error rate is proposed.Since it has been reported that the ensemble strategy is an effective technique to improve the stability in the field of feature selection,our algorithm will then design an ensemble selector to evaluate the significance of attribute,a set of the fitness functions instead of only one fitness function is used.The experimental results show that the ensemble approach is effective in improving the stabilities of both reducts and classification results.In order to verify the effectiveness of this method in different fuzzy rough set models,four different fuzzy rough classifiers are used as the third-party classifiers,and we also can achieve higher classification accuracy.(3)In fuzzy rough set model.Classification consistency is designed with the help of joint distribution matrix.Furthermore,a new attribute criterion evaluation mechanism is proposed,which takes the harmonic average of classification consistency and classification performance as a new attribute criterion evaluation mechanism.The experimental results show that the harmonic average approach can not only improve the decision consistencies and the classification accuracies significantly,but also bring us more stable reducts.