Function Mining Based on Gene Expression Programming ——Convergency Analysis and Remnant-guided Evolution Algorithm

Mining the interesting function from the very complex and large data sets is an important research direction in DM and KDD. Traditional methods depend on experience and professional knowledge and involve subjectiveness and blindness when determine function type and can not work well with complex function expression. Function mining method basted on Genetic Programming (GP) can overcome the shortcomings of the traditional methods. But the GP method is not efficient. Gene Expression Programming (GEP) is a new function mining method. The convergency property of function mining based on GEP was studied and a revised algorithm of GEP—Remnant-Guided Evolution Algorithm (RGEA) was proposed.The experiments over Genetic Programming (GP), GEP and RGEA showed that (a)all algorithms find the target function from data with low noise; (b) the convergency speed of GEP is 20 times faster than GP,and RGEA is 60 times faster than GP; (c)For very complex data with unkown function type,GEP and RGEA are respectively 900 and 1800 times faster than GP.