Quick reduct with multi-acceleration strategies in incomplete hybrid decision systems

Hao Ge, Chuanjian Yang, Yi Xu, Gongjian Peng

Abstract: In the era of big data, discovering knowledge from incomplete and mixed complex systems is an important research topic that has always been the focus and favor of researchers. Attribute reduction is one of the important research contents in the fields of knowledge discovery, data mining and etc. For improving the performance of attribute reduction, some scholars have proposed many approaches and acceleration mechanisms. As an effective data processing model, the matrix has some bottlenecks in obtaining the reduct efficiently, and there is less research on its accelerated reduction strategies and algorithms. Therefore, the main motivation of this article is to explore and apply various acceleration strategies in matrix reduction methods to improve the efficiency of knowledge discovery in incomplete hybrid decisions systems (IHDSs). First, for incomplete mixed data, the combination tolerance relation is presented. Based on the combination tolerance relation, the corresponding operations and properties of the related matrix are studied, and a matrix-based positive region reduction method is proposed. Next, we develop the incremental computing strategy for updating the matrices to avoid re-calculation and the gradually reducing processing scale strategy for compressing the processing volume in the reduction process; then, two types of matrix-based heuristic reduction acceleration algorithms are designed to obtain a reduct of IHDSs. Finally, some experiments are carried out on nine typical data sets in 15 UCI data sets to verify the effectiveness and efficiency of proposed algorithms.