Attribute Reduction Function Mining Algorithm Based on Gene Expression Programming
Changan Yuan, Changjie Tang, Jie Zuo, Chen An-long, Yuanguang Wen · 2006
When mining non-linearity function with large number of variables, traditional methods cannot effectively reduce the conditional attributes. To solve the problem, this paper proposes GEP-ARFM model. The model includes the concepts of marginal gene, marginal fitness, and revised fitness and the algorithms of GEPAMF, GARFM-GEP, and SARFM-GEP. The comparison experiments show that (1) both GARFM-GEP and SARFM-GEP can effectively reduce the conditional attributes to find the best function expression. (2) The precision of function expression by using SARFM-GEP is approximate with using GARFM-GEP algorithm. (3) SARFM-GEP method is 300 times faster than GARFM-GEP in the case of 20 independent variables. (4) The fitness value of the function expression got by using GEP-ARFM model is 24.6% greater than the traditional method