The Research Of Classification Rules Mining Based On Multi Artificial Fish Warm Cooperation Algorithm

Artificial Fish Warm Algorithm (AFWA) is an emerging technology in the entire field of artificial intelligence. It is often applied to various optimization problems and data mining since2002when it’s presented for the first time. The process of data classification is that a sample data set should be identified into the appropriate categories according to their attribute characteristics. As one of the most important content of data mining, it has been widely applied to various fields. Extracting classification rules is the further performance of data classification; it is aimed chiefly at predicting the category of a sample data. So, the classification rules can be seen as a kind of significant DM knowledge. There is little research about AFWA’s application in data classification and extracting classification rules at present, what is the core research of this paper.A multi artificial fish swarm cooperation algorithm (MAFWA) that is inspired by the ideology of multi-swarm’s co-evolution is proposed in this article, and is applied to the study of extracting classification rules. The main idea of this method are that:multi artificial fish swarm will be designed and initialized, each artificial fish swarm is aimed at extracting classification rules for a category, multi swarms will evolve simultaneously to get the integrated classification rules. MAFWA can reduce the time of sophisticated communications between each two artificial fish and the complexity of this algorithm by the comparison with traditional single artificial fish swarm algorithm, that extract classification rules for all categories. The method of multi-swarm’s co-evolution can obviously speed up the convergence of AFWA and improve the accuracy of classification rules.In order to make up for the low diversity degree of artificial fish swarm and plunging into a local optimal solution, the self-adapting selecting operator, cross operator and mutation operator from GA are adopted to improve the basic AFWA, a multi artificial fish warm algorithm with cross, mutation And Choose (MAFWA_CMC) is proposed further, that can avoid premature convergence and have a better performance.Finally, the simulation experiment results show that these methods are effective. MAFWA generated the classifying rules with high accuracy fastly, and the efficiency and precision of MAFWA fully exceeds the traditional single artificial fish swarm algorithm. The convergence speed and accuracy of classification rules of MAFWA_CMC is evidently superior to MAFWA, but its computation time is greater than MAFWA because of the adoption of selecting operator, cross operator and mutation operator. But the disadvantage can be offside in the process of classification of high-dimensional data by the fast convergence speed. The comparison of MAFWA_CMC and multi particle swarm optimization (MPSO) is also performed; the results show that the performance of improved MAFWA_CMC is slightly better than MPSO.