A Fuzzy Clustering Algorithm Based On Parameter Optimization

Cluster analysis is an important branch of data mining. The introduction of fuzzytheory brings new vitality into this problem. Now, fuzzy cluster analysis has beenwidely applied to statistics, marketing, biological and other areas. Most clusteringalgorithms are not applicable to datasets with severe outliers inside, in order toovercome this shortcoming, we propose a fuzzy clustering algorithm based onparameter optimization. The main contents of this paper are as following.Recent modifications of Fuzzy c-means(FCM) using Lp-norm distances increaserobustness to outliers, but how to choose suitable p parameters is still a difficult task.We present an improved algorithm named FMMLE(Fuzzy Multi-Metric LocationEstimation) to solve this problem. Various experiments show that the main advantage ofthis algorithm is that it allows for the adjustment of parameter p according to the datafeature, which makes it robustness. In addition, based on FMMLE, we propose anoutlier detection algorithm named FRMMC(Fuzzy Robust Multi-Metric Clustering). Inorder to testify the effective of FRMMC, simulation is done based on synthetic data andWisconsin Breast Cancer data. Finally, we applied this parameter optimization methodto nonlinear classification problems, and propose a algorithm FKMMLE(Fuzzy KernelMulti-Metric Location Estimation), which uses simulated annealing method to searchthe optimal kernel parameters. Simulation results demonstrate that FKMMLE isefficient and robust.