一种基于马氏距离的可能性聚类方法

Mahalanobis Distance-Based Possibilistic Clustering Algorithm and Its Analysis

可能性聚类方法在数据分析和模式识别领域被广泛应用。本文通过选择马氏距离,构造一种特殊的基于马氏距离的可能性聚类方法。该方法在保持可能性聚类性能的同时,能有效地防止一致性聚类的发生,依照最大最小概率原理,根据聚类的结果直接推断出聚类结果的误分下界,从而判断聚类的有效性。最后通过图像分割实验和标准数据集实验,验证了该算法的优越性。

The possibilistic clustering algorithm is used in data analysis and pattern recognition.By Mahalanobis distance with the algorithm,a special Mahalanobis distance-based possibilistic clustering algorithm is proposed.This algorithm has the advantage of possibilistic clustering.Meanwhile,it can prevent leading to a single cluster center and has a distinctive characteristic,thus immediately deducing the lower bounds of the obtained clustering results.Based on the principle of the minmax probability,it can decide whether the clustering is effective or not.Finally,experiments on image segmentation and UCI data show that the algorithm is effective.