一种新的混合球壳形数据聚类方法

A New Hybrid Approach for Spherical Shell Clustering Data

模糊C-球壳聚类算法(FCSS)广泛应用于模式识别与机器学习领域。传统FCSS算法采用梯度法和交替寻优策略来求解,因此对初始值比较敏感,往往只能得到模型的局部极值点,从而影响聚类的效果,甚至使所采用的方法失效。将粒子群优化算法(PSO)、遗传算法(GA)与传统FCSS算法相结合,得到一种有效的混合球壳聚类算法PSO-GA-FCSS,数值实验结果表明新算法的性能优于FCSS算法、遗传算法与FCSS相结合的算法GA-FCSS以及粒子群算法与FCSS相结合的算法PSO-FCSS。

The fuzzy C-spherical shells（FCSS） algorithm is widely applied to pattern recognition and machine learning.The classical clustering algorithm is based on gradient method and alternative optimization strategy;its disadvantages are sensitive to the initial values and easy to trap into a local optimal solution,affecting the clustering effect,even causing a false result.An efficient hybrid spherical shell clustering algorithm called PSO-GA-FCSS is proposed in this paper,which is based on the combination of FCSS,particle swarm optimization（PSO） and genetic algorithm（GA）.As shown by the results of the computational tests,the clustering quality is satisfactory.The simulation results show that the performance of the new algorithm is better than FCSS,GA-FCSS and PSO-FCSS.