The Function Mining Based On GEP And Its Application In Data Filter

Data mining is a process that discovers knowledge from mass data and information. It plays an important role in decisions making and actions as guidance. As a method of data mining, discovering mathematical relations from experimental data is always an important task in scientific studies.Gene Expression Programming (GEP) is a new member in the family of Genetic Algorithm (GA). It is different from traditional GA in expressing and processing of individual and form of result. GEP is used as the main algorithm of our function mining method due to its advantages in expression string handling.The real data collected from a system always include error because of noises. In order to reduce the mean square error of the data, a new filtering method is proposed that filters data .according to function discovered from the data set.The main contribution of this work includes:(1) Analyzing the specialties and difficulties of function mining, which is a special mode of Data Mining.(2) Introducing an efficient and robust function mining method GEPFM (GEP based Function Mining). Through GEP based algorithm UEM and MEM, GEPFM can find the expression of the goal function, and effectively handle thedomain split of functions.(3) Implementing an interactive GEP Experiment Platform named GEPxP by C++ on Windows2000. This .system can perform experiments of function mining and data filtering in diverse parameter configurations.(4) By extensive experiments on GEPxP, it is shown that the new function mining method has good performance. With high mutation frequency, the success rate of mining is 20%~80%, within less than 10 seconds. In the data filtering test, the mean square of sample data can be greatly reduced and it is sometimes near 0.The contents of this paper are organized as following:Chapter 0 introduces the importance of the study of function mining. Chapter 1 introduces the concept, process and catalog of data mining. Chapter 2 first analyzes the specialties and difficulty of function mining, and compare GEP with traditional GA, then introduces the function mining method UEM and MEM and present them in detail, and finally demonstrate the complexity of the methods. Chapter 3 analyzes a - B Filter which is used in some dynamic systems, introduces the function mining based data filter which is named f-Filter. Chapter 4 discuses the system architecture, interface, and the implementation of some module of the GEPxP are presented in detail. Chapter 5 gives experiments results, and evaluates the performance of GEPxP in different configurations. Chapter 6 gives conclusion and discusses prospects for the future works.