Research On Clustering Algorithm Baesd On Group Search Optimization

Clustering algorithm as an unsupervised learning algorithm, divide data set into several classes by the similarity measure without prior knowledge of data distribution.Through clustering, we find the correlation between data samples. However, conventional clustering algorithms perform poorly in terms of algorithm performance. Therefore, it is necessary to improve clustering performance and robustness by using different parameters and heuristic function rules optimized by group intelligence. Besides, group search optimization algorithm has many advantages, such as few parameters, simple operation and not easy to fall into local optimum. In this paper, the ideas of group search optimization are incorporated into the clustering analysis algorithm, and conventional clustering problem is transformed into optimization problem. Four new clustering algorithms based on group search optimization are proposed. Mainly done the following work:1. The conventional Group Search Optimizer (GSO) is not free from some drawbacks such as easily falling into local optimum, a relative long computing time and lower convergence accuracy. This paper proposes a differential ranking-based group search optimizer (DRGSO) algorithm to alleviate these limitations. There are mainly two improvements in the design of DRGSO. First, the population is initialized according to the ranking of fitness values. With this regard, the population obtains heuristic information and alleviates premature convergence to some extent. Second, four evolutionary operators based on differential strategies are constructed to improve the convergence of the algorithm and enhance the population diversity. To demonstrate the performance, eleven benchmark functions are included to evaluate the performance of DRGSO. Experimental results indicate that the proposed DRGSO exhibits better performance in comparison with the GA,PSO and GSO in terms of accuracy and speed of convergence.2. This paper proposes four clustering algorithms based on group search optimization,namely GSO clustering algorithm, GSO clustering algorithm based on average, DRGSO clustering algorithm and DRGSO clustering algorithm based on average. On the one hand,population structure is used to encode the positions of cluster centers in order to optimiaze the cluster assignment process. On the other hand, according to the range of the producers'fitness, the local mean value strategy is proposed to avoid getting into the local optimum and improve the sharing pattern of information resources among individuals.3. Using the differential mutation operator model, the cluster moves more diversity and improves the global searching ability of the clustering algorithm. The experimental results of the Iris and Wine data set in the UCI database show that the clustering algorithm of group search optimization has more obvious clustering effect and better stability and robustness.