Data-driven Approach To Acquisition Of Variable Precision Threshold In Variable Precision Rough Set Model

Rough set theory is a valid mathematical theory to deal with imprecise, uncertain, and vague information. Without domain knowledge, rough set theory can deal with uncertainty knowledge, using certain methods. It has been successfully applied in such fields as machine learning, data mining, intelligent data analyzing and control algorithm acquiring, etc.However, there is usually noise data in real life databases, especially in huge databases. It is very difficult for classical rough set theory to deal with these data. Variable precision rough set, proposed by Canadian Professor Ziarko.W in 1993, is an extension of classical rough set theory, which could deal with noise data, by introducing a variable precision thresholdβ. The variable precision rough set theory succeeds in processing noise data in the past years.The difference between the classical rough set theory and the variable precision rough set theory is the variable precision thresholdβ, which affectsβpositive area,βboundary area, andβnegative area. Obviously, the variable precision thresholdβcan greatly affect the capability of knowledge acquisition of the variable precision rough set theory. However, the variable precision thresholdβneeds to be set up by domain experts, which needs prior domain knowledge. Thus, the variable precision rough set theory could not be applied if domain prior knowledge is not available, a data driven approach is proposed to implement the self-learning acquisition of the variable precision thresholdβ. The experiment results show that the variable precision thresholdβacquired by this data-driven approach can improve the capability of knowledge acquisition of the variable precision rough set theory. Due to the impact of the variable precision thresholdβ, the situations of attribution reduction based on the variable precision rough set theory are much more complicated than the ones based on the classical rough set theory. Abnormal reduction may be generated. The nature of the attribution reduction is granular combination. In this paper, from the view of granular combination, the situations and the reasons for the occurrence of the abnormal cases are analyzed. A data-driven approach to attribution reduction based on the variable precision rough set theory is also proposed.