Attribute Reduction Based on the Multi-annulus Model

Attribute reduction is one of the important methods in data mining preprocessing. In order to reduce the time consumption of attribute reduction, from the perspective of data distribution, the multi-annulus model is established for samples under different labels. The multi-annulus model can adaptively generate multiple annuluses with different radii according to the different data densities, and the intersection between annuluses is regarded as the boundary domain of annuluses, which introduces a new angle to solve the attribute reduction problem. The key to solving the attribute reduction of the multi-annulus model is to take the quality in the annulus as the metric criterion. Additionally, the radius ratio of the annulus and the boundary region of the annulus is used as the weight of the quality in the annulus. To obtain the change of the quality in the annulus caused by each candidate attribute added to the attribute reduction pool, the forward greedy strategy is employed. Its essence is to divide the unevenly distributed data into corresponding annuluses and reduce the preprocessing process of the data itself, so as to accelerate the process of attribute reduction. Compared with the other three mainstream attribute reduction algorithms, the final experimental results show that the proposed algorithm can greatly shorten the time of attribute reduction by introducing the multi-annulus model.