%0 Conference Paper %1 Li:gecco05lbp %A Li, Xin %A Zhou, Chi %A Xiao, Weimin %A Nelson, Peter C. %B Late breaking paper at Genetic and Evolutionary Computation Conference (GECCO'2005) %C Washington, D.C., USA %D 2005 %E Rothlauf, Franz %K algorithms, expression gene genetic programming programming, %T Prefix Gene Expression Programming %U http://www.cs.uic.edu/~xli1/papers/PGEP_GECCOLateBreaking05_XLi.pdf %X Gene Expression Programming (GEP) is a powerful evolutionary method derived from Genetic Programming (GP) for model learning and knowledge discovery. However, when dealing with complex problems, its genotype under Karva notation does not allow hierarchical composition of the solution, which impairs the efficiency of the algorithm. We propose a new representation scheme based on prefix notation that overcomes the original GEP's drawbacks. The resulted algorithm is called Prefix GEP (P-GEP). The major advantages with P-GEP include the natural hierarchy in forming the solutions and more protective genetic operations for substructure components. An artificial symbolic regression problem and a set of benchmark classification problems from UCI machine learning repository have been tested to demonstrate the applicability of P-GEP. The results show that P-GEP follows a faster fitness convergence curve and the rules generated from P-GEP consistently achieve better average classification accuracy compared with GEP

@inproceedings{Li:gecco05lbp, abstract = {Gene Expression Programming (GEP) is a powerful evolutionary method derived from Genetic Programming (GP) for model learning and knowledge discovery. However, when dealing with complex problems, its genotype under Karva notation does not allow hierarchical composition of the solution, which impairs the efficiency of the algorithm. We propose a new representation scheme based on prefix notation that overcomes the original GEP's drawbacks. The resulted algorithm is called Prefix GEP (P-GEP). The major advantages with P-GEP include the natural hierarchy in forming the solutions and more protective genetic operations for substructure components. An artificial symbolic regression problem and a set of benchmark classification problems from UCI machine learning repository have been tested to demonstrate the applicability of P-GEP. The results show that P-GEP follows a faster fitness convergence curve and the rules generated from P-GEP consistently achieve better average classification accuracy compared with GEP}, added-at = {2008-06-19T17:35:00.000+0200}, address = {Washington, D.C., USA}, author = {Li, Xin and Zhou, Chi and Xiao, Weimin and Nelson, Peter C.}, biburl = {https://www.bibsonomy.org/bibtex/2ff541900dbc3e575def3350b3b2d16c9/brazovayeye}, booktitle = {Late breaking paper at Genetic and Evolutionary Computation Conference {(GECCO'2005)}}, editor = {Rothlauf, Franz}, interhash = {0403d8e904134aebe744add50822ce75}, intrahash = {ff541900dbc3e575def3350b3b2d16c9}, keywords = {algorithms, expression gene genetic programming programming,}, month = {25-29 June}, notes = {Distributed on CD-ROM at GECCO-2005}, timestamp = {2008-06-19T17:45:33.000+0200}, title = {Prefix Gene Expression Programming}, url = {http://www.cs.uic.edu/~xli1/papers/PGEP_GECCOLateBreaking05_XLi.pdf}, year = 2005 }