Fuzzy Rough Based Feature Selection by Using Random Sampling

Feature selection, i.e., Attribute reduction, is one of the most important applications of fuzzy rough set theory. The application of attribute reduction based on fuzzy rough set is inefficient or even unfeasible on large scale data. Considering the random sampling technique is an effective method to statistically reduce the calculation on large scale data, we introduce it into the fuzzy rough based feature selection algorithm. This paper thus proposes a random reduction algorithm based on random sampling. The main contribution of this paper is the introduction of the idea of random sampling in the selection of attributes based on minimum redundancy and maximum correlation. First, in each iteration the significance of attribute is not computed on all the objects in the whole datasets, but on part of randomly selected objects. By this way, the maximum relevant attribute is chosen on the condition of less calculation. Secondly, in the process of choosing attribute in each iteration, the sample is different so as to select the minimum redundancy attribute. Finally, the experimental results show that the reduction algorithm can obviously reduce the running time of the reduction algorithm on the condition of limited classification accuracy loss.