Research On Gene Expression Programming And Its Application In Function Mining

Nowadays there is a rapid development of the society and the corresponding volume ofsocial information. However, we are trapped in a situation of“rich information but poorknowledge”. Function mining, as a important branch of data mining, is aimed to discover thefunctional model which reflect existing rules and its variation tendency and further directpractical work. This paper is mainly focused on the Gene Expression Programming and itsapplication in the field of function mining. Gene Expression Programming (GEP), proposed by aPortuguese named Ferreira, is a new type of self-adaptive evolutionary algorithm which is basedon and develops from the advantages of genetic algorithm and genetic programming. It is a newknowledge discovery technology. At present, GEP has been successfully applied to the field offunction mining, prediction of time series, machine learning, image processing and associationrules mining. Compared with traditional methods, GEP enjoys great advantage by tacklingcomplicated problems with simple coding.This paper is mainly focused on study of the GEP and the realization of function miningthrough GEP. It comprise of the following parts.Firstly, the paper summarize the basic features, general situation of development andapplication background of the data mining particularly function mining. It lays great emphasis onthe description of the core technologies of GEP which refers to the unique coding form andmethod of operation which mainly includes the chromosome compositions of GEP and frameworks, computational methods of fitness function which aims to evaluate the adaptive capacityof the solution, varied genetic manipulation such as selection, mutation and special crossover ,insertion sequence transposition, root insertion sequence transposition, gene transposition,one-point recombination, two-point recombination, gene recombination and numeric constantand basic procedure of GEP etc further analyze the advantages which GEP enjoys over otheralgorithms.Secondly, GEP has its disadvantages and needs developing and perfecting. On one hand,astringency of GEP algorithm should be guaranteed to maintain the variety o f the population, onthe other, the effort to force the solution to close to the optimum direction will definitely affectthe variety of population thus trapped in situation of local optimum which will cause theprematurity. This paper points out the disadvantages and according causes of the traditional GEPin maintaining variety of population and increasing efficiency of algorithm, moving forward todesign the improvement program which covers following parts, restraining premature through reverse elimination strategy, automatic fallback mechanism based on sufficiency, recombiningstrategy of derivative coding, increasing space-time efficiency through no-tree decoding modeand increasing fitness of function model through a determining method of numeric constant.Thirdly, in practice, this paper highlights the specific methods and key points of GEPtechnology when applying to function mining particularly the specific methods which combinethe two effectively such as the issue concerning the decoding of function expression, selection offitness function, operation of genetic manipulation and the specific process of the realization offunction mining, further reflect the feasibility and unique advantage when GEP is applied tofunction mining.Finally In order to verify the application effect, a software of function mining through GEPis programmed and unitary, binary and polynary function mining experiments are undertaken. Inaddition, a prediction model is established for the vehicle population of Shandong Province, allof these has provided reliable experimental basis for the study. Meanwhile, through thecomparison between the methods stated in this paper and the traditional function mining as wellas the basis GEP algorithm, modified GEP algorithm is proved to have higher accuracy andfitness regarding to function model, wider sphere of application and better performance ofprediction and direction. At last, a conclusion of the paper is made foreseeing the futuredevelopment and further study work.