Research On Rules Generation Algorithm Based On Rough Sets

Rough set algorithm is a new algorithm in data mining domain. It is good at dealing with fuzzy data, but the core algorithm-reduct is too complex to apply. This is why rough set algorithm isn't used widely. The aim of this paper is we want to reduce the complexity of rough set rules generation algorithm, but keep its classification capbility. For this, this paper first time provides a new rules generation algorithm SFRGA (Short First Rules Generation Algorithm ) which is based on the classic rough set theory. Comparing whith the classic algorithm, classic algorithm needs reduct, but our algorithm doesn't; the classic algorithm is based on partition, SFRGA is based on covering. This paper proves that SFRGA has the following properties: 1) SFRGA can make the same data covering rate as the others; 2) SFRGA can generate less rules than the classic algorithm; 3) the time complexity of SFRGA is lower than the classic algorithm, and as the data grows, it's more obvious; to compare the rules'complexity, this paper provides a new standard MRL (Mean Rules Length), and proves 4) the MRL of SFRGA is always lower than the classic one. Besides the proof, this paper also implements SFRGA by haskell language, and the related experiments proves its good property too. SFRGA is not only innovation of rough set theory, but also performs good at rules generation application.