The Study For Mining Classification Rules Based On Genetic Algorithms

Data Mining is a domain which tries to extract knowledge and interesting information from very large-scale databases. This knowledge is hidden, unknown, but potentially useful. Mining classification rules is a procedure to construct a classifier through studying the training dataset. It is a very important part of Data Mining and Knowledge Discovery. In essence, the goal of data mining is to discover knowledge which is highly accurate, comprehensible and interesting. This paper has done some research on classification based on GA. We also do a further research on constructing comprehensible classification rules using niche-based GA and finding interesting classification rules using adaptive GA.Genetic Algorithm is an overall random searching method based on the theory of evolution and molecular genetics. In this paper we study the key technology to the implementation of genetic algorithm and put forward a particular scheme of how to determine the parameters and operations when classify, including individual encoding, fitness function design, GA operators design, etc. Thus, we give the method of how to mining classification rules using GA in theory. Furthermore, in order to assure that the acquired classification rules to be accuracy as well as comprehensible, we put forward a method to compute the comprehensibility of the classification rule using attribute information gain, different to the other methods which evaluate the comprehensibility of the rule only by its simplicity. Thus the output of the mining is more understandable and informational. We also do it using a niche-based GA. The experimental results indicate that it is a feasible method, which is easy and adaptable to the evaluation of the performance of genetic algorithms. Currently, the discovery of interesting knowledge remains a challenge for data mining algorithms. This paper put forward a GA-based method to discover interesting classification rules. First, we design the fitness function which can evaluate the interestingness of a rule by using attribute's information gain and by setting the weight of the information gain. It is different to other methods in that it combine the subjective and objective interestingness-measure methods together other than separate them. Then, as it is desirable to reduce the restrained speed of the GA, we discuss mining classification rules with adaptive GA. Experiments show it can find interesting classification rules in the database.