In the era of big data, massive amounts of ordered data exist in all walks of life. Incremental attribute reduction of ordered data has become a widely studied topic, which can identify essential attributes and reduce the dimensionality of attributes, as well as improve the classification ability of learning models. However, most of the existing incremental attribute reduction methods can only handle ordered decision system (ODS) with one-dimensional variations, and the application scenarios are relatively ideal. Therefore, in response to this situation, this paper introduces an incremental attribute reduction method based on the dominance-based rough set approach (DRSA) to effectively handle multi-dimensional variations of ODS. First, this paper uses self-information as an uncertainty measure, which can simultaneously consider certain and possible classification information to more accurately capture the preference relation between attributes and decisions. Secondly, the matrix calculation process is optimized by processing the dominance-based relation tiangular matrix (tDRM), which greatly saves time. In addition, a more efficient generalized decision (GD) method is introduced to calculate the upper and lower approximation, which further improves the computational efficiency of the algorithm. Finally, the verification on multiple datasets shows that the proposed algorithm can effectively remove redundant attributes and greatly save the reduction time.
Read full abstract