Optimization Approaches And Its Applications For Attribute Reduction

Dimensional disaster is one of the common challenges in machine learning and data mining technology.Attribute reduction,as a feature selection technology,chooses the smallest subset of attributes as far as possible to accurately describe learning data.The existing problems of most attribute reduction algorithms can be summarized as follows:(1)The local optimal solution obtained by attribute reduction is unstable,which often leads to ambiguity;(2)The evaluation mechanism of attribute criteria is not perfect at present;(3)The research on the intrinsic structure of data and the relationship between attributes and labels is insufficient.In view of this,the methods of the ensemble strategy,multi-criteria based attribute reduction and local view to optimize attribute reduction algorithms are proposed,respectively.The contents and innovative achievements of this study are mainly as follows:(1)An ensemble significance based attribute reduction approach is proposed.The experimental results obtained by utilizing the neighborhood rough set method show that the new approach not only obtains a more stable reduct,but also attains the classification results with high uniformity;(2)Classification consistency is designed with the help of joint distribution matrix in ensemble learning.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 multi-criterion based attribute reduction can not only improve the decision consistencies without decreasing the classification accuracies significantly,but also bring us more stable reducts;(3)A local view based cost-sensitive attribute reduction is introduced.It fully considers the change of decision-making cost of each object rather than the change of decision cost on the region.Furthermore,the local view based attribute evaluation mechanism of decision cost is applied to heuristic algorithm to compute reduct.the experimental results illustrate that by comparing the traditional reduction,the proposed new approach can decrease decision cost effectively,which can select a better subset of attributes.