Research on Attribute Reduction Method Based on Local Dependency

Attribute reduction is one of the research hotspots in the field of data mining. Although the result of attribute reduction algorithm based on single attribute identification matrix is better, it is still not efficient enough to deal with large-scale information system problems. In this paper, the concept of sub matrix of single attribute identification matrix is proposed. Based on the sub matrix, the calculation method of local dependency degree is given, and an attribute reduction algorithm based on local dependency degree is designed. If the equivalence class of information system is regarded as basic knowledge granules, this algorithm first finds an attribute set to separate the first particle from other particles, and then adds attributes to the attribute set in order to separate the second particle from other particles. Repeat the above operation until all particles are distinguished, and the resulting attribute set is called reduction set. This algorithm reduces the time and space complexity of reduction algorithm to a certain extent. The effectiveness of this method is verified by UCI data set, which provides a method for attribute reduction.