GEP-NFM: Nested Function Mining Based on Gene Expression Programming

Mining the interesting functions from the large scale data sets is an important task in KDD. Traditional gene expression programming (GEP) is a useful tool to discover functions. However, it cannot mine very complex functions. To resolve this problem, a novel method of function mining is proposed in this paper. The main contributions of this paper include: (1) analyzing the limitations of function mining based on traditional GEP, (2) proposing a nested function mining method based on GEP (GEP-NFM), and (3) experimental results suggest that the performance of GEP-NFM is better than that of the existing GEP-ADF. Averagely, compared with traditional GEP-ADF, the successful rate of GEP-NFM increases 20% and the number of evolving generations decrease 25%.