Application Of The Hybrid Algorithm Of Fuzzy C Mean And Genetic Algorithm

Cluster analysis is a widely used data mining technique which is commonly found in a variety of statistical software(such as SPSS) and data mining tools. Among all the approaches, partition based algorithms like k-Means algorithm and FCM algorithm are the most popular ones.FCM algorithm introduces the fuzzy theory into clustering algorithms which gives us the capacity of dealing with data generated by real-life. This paper first introduces the basic principles of FCM algorithm, time and space complexities, and then analyzes the shortcomings of FCM algorithm. The biggest flaw of FCM algorithm is its sensitivity of the initial cluster centroid, it may lead to not able to get the best clustering results. The most common way to overcome this drawback is to take multiple sets of parameters and select the best result, but the results are very difficult to predict.In order to overcome the shortcomings FCM algorithm, this paper introduces genetic algorithm to find the optimal initial conditions by the global optimization capability of genetic algorithm. Based on genetic algorithm, we combine the FCM algorithm with genetic algorithm as an improved FCM algorithm. And in order to handle the high computational complexity caused by the genetic algorithm, sampling algorithm was also introduced to improve the scalability.In order to verify the effectiveness of the improved method, we use a standard data set with FCM algorithm and the improved method as control. In order to test the effectiveness of the improved algorithm on a real problem, we also get some real data from a certain company’s open platform.In conclusions, there are some points achieved in this paper can be summarized as follow:(1) In this paper, an improved method of FCM algorithm based on genetic algorithm is came up with.(2) Due to the great computing resource need of genetic algorithm, a sampling method is used to reduce the computation.(3) Experimental data indicates that the proposed improved method ensures the convergence of the algorithm to the global optimum and it has a clear advantage on reducing the number of iterations of FCM.(4) At the same time, the application to the real data can clearly characterize clusters, and they are easily to be interpreted with information and knowledge.Although our research result has good performance in certain applications, it’s under the circumstance of small data set and we have to give the k parameter manually. If we want to apply our result into complex problem, we need further research and experiment.