Application Of Integrative Algorithm Of Feature Weighted FCM Clustering And PSO Algorithm

In recent years, the fuzzy cluster analysis which is widely used in numerous kind of subjects study is an important research method. It has been valued in many fields such as economic, financial, life science, medical diagnosis, business management, geology, astronomy and so on. Especially, it plays a very important role in the multivariate analysis and image recognition. Among many fuzzy clustering methods, the Fuzzy c-means(FCM) has become the most popular method due to its simple process and remarkable effectiveness. However, Fuzzy c-means clustering method also has its own weakness: FCM method is especially sensitive with the initialization of clustering prototypes(or partition matrix) and easy to fall into local optimum solution and the traditional FCM algorithm ignores the different contribution of different features. To solve these problems above, we hope to seek an improvement on FCM algorithm.In numerous optimization algorithms, particle swarm optimization(PSO) has become the most popular one because of its satisfactory versatility and simple process. PSO algorithm is a kind of evolution algorithm based on multidimensional randomly searching and acceleration to obtain the global optimization, which is applied to many fields. And the inertia weight plays a important role in this algorithm.In this paper, we take the parameters, inertia weight and learning factory, of particle swarm optimization algorithm in adaptive processing, adjusted dynamically, and take a treatment of feature weight on objects in the fuzzy c-means algorithm in view of the shortcoming of fuzzy c-means algorithm that it does not consider the different influence of different features. Eventually, we integrate this feature weighted fuzzy c-means algorithm with the adaptive inertia weighted particle swarm optimization algorithm to create a new hybrid algorithm, and then, strive to make up for the inadequacy of traditional FCM algorithm, and obtain better clustering results. Finally, we apply this hybrid algorithm to real datasets, and explain the advantages of the algorithm proposed in this paper over the traditional algorithms according to the experiment results.