Research On Automatic Clustering Algorithms Based On Particle Swarm Optimization And Its Applications

As one of important tasks of data mining, clustering analysis is applied in many areas like management, medical science, economics and biology. The clustering, which can determine the cluster number automatically, has very important significance in practice. Clustering can be regarded as an optimization problem. So, this paper aims to deal with automatic clustering problem using particle swarm optimization. First, this paper applies simple particle swarm optimization to optimize automatic clustering index function and finds out all cluster centers and cluster number. Second, niche particle swarm optimization is introduced and improved in this paper to optimize the density function. At last, the two automatic clustering algorithms are used in customer relationship management to do the customer segmentation. The empirical results indicate that the methods proposed by this paper have better practical value. The main research work in this paper is as follows.(1) This paper aims to propose an automatic clustering algorithm based on k-harmonic means and particle swarm optimization. This method implies a new kind of encoding to determine the optimal cluster number. Furthermore, k-harmonic means is used during the process of optimization to accelerate the convergence. Compared to k-means, k-harmonic means is independent on the initialization of cluster centers, which makes the algorithm stable and robust.(2) This paper puts forward another automatic clustering algorithm which introduces niche particle swarm optimization (NichePSO) into clustering analysis and applies NichePSO to optimize density function. First, this paper improves particle training model from main group and increases the ability of their space search. Second, the radius of subgroup is defined adaptively according to the actual clustering problem, which can be useful for the form and search of niches. At last, a new method that distributes samples to the corresponding cluster is brought. Numerical results illustrate that this algorithm based on density function and NichePSO could cluster the unbalanced density data into the correct clusters automatically and accurately.(3) This paper applies the two automatic clustering algorithms into commercial data and does research on customer segmentation of customer relationship management. Customer segmentation is the core part and base of customer relationship management. This paper chooses some auto repair records from4S stores and refines the objective dataset based on RFM index system. Then the two automatic clustering methods are used to cluster the objective dataset and partition it into several smaller groups. Finally, this paper analyzes every group based on its attributes, finds some substantial rules and proposes the corresponding marketing strategies according to the operation of4S. The results indicate that the two automatic clustering methods can cluster practical commercial data reasonably and have good performance in commercial application.