Weka平台上解决聚类的改进差分进化算法

针对K均值算法的缺陷,提出一种用于解决聚类问题的差分进化算法对聚类的准则函数进行优化,为了能够进一步增强算法的全局搜索能力,引入一种基于种群适应度方差的自适应策略来动态调整变异概率CR和规模因子F等参数,充分利用在Weka工具中的类和接口,并将新提出的算法嵌入到平台中.在Weka平台上将该算法与K均值算法在3个UCI数据集上进行比较.仿真实验结果表明,该算法能够有效克服K均值算法的缺陷,能够获得较高的聚类质量.

Advanced differential evolution algorithm for clustering on Weka

After analyzing the drawbacks of the K-means algorithm,a novel differential evolution algorithm for solving clustering problem is proposed to optimize the criterion function for clustering.In order to further enhance the capability of global search,a self-adaptive strategy using fitness variance of the population is introduced to adjust scaling factor and crossover probability.The proposed approach is implemented on the Weka platform where its classes and interfaces are fully utilized.At last,the proposed algorithm is tested on three UCI datasets and compared with K-means.The simulation results indicate that the proposed algorithm can acquire better clustering performance.