基于样本与特征双加权的自适应FCM聚类算法

Adaptive FCM clustering algorithm based on sample and feature weights

针对传统FCM算法无法获得令人满意的聚类结果的问题,提出了基于样本与特征双加权的自适应FCM聚类算法。采用特征和样本双加权的策略,以特征权重信息熵作为代价函数,与样本权重、特征权重相融合,通过迭代优化的方法动态计算各属性特征对不同类别的权重系数、每个样本对聚类的重要性权重值,综合考虑各个样本的贡献度和各个特征的重要性,从而达到提高聚类结果质量的目的。使用5个来自UCI的标准机器学习数据集,对聚类算法的有效性进行验证。结果表明,对于具有不同样本贡献度和不同特征重要性的数据集,提出的算法具有较好的聚类效果。

As a classical machine learning algorithm, FCM （fuzzy C-means ） clustering algorithm is widely used in many applications. For datasets with different feature importance and different sample con- tribution, features weight based algorithms are proposed to improve the clustering quality in some degree, but can not obtain the optimal result. Based on the weighting strategy of feature and sample, the cost function of feature weight information entropy, and the combination of sample weight, attribute feature weight and objective function, an adaptive FCM clustering algorithm with both sample and feature weight is proposed, by adopting the iterative optimization to measure the weighting coefficients of each sample feature upon each cluster dynamically, as well as the importance of every sample to the clustering. The algorithm takes full account of the influence of different feature and sample contributions on the clustering results, so as to achieve the purpose of improving the clustering effect. The clustering algorithm is veri- fied by using a set of test data sets from the UCI standard machine. The experimental results show that the proposed algorithm has better clustering accuracy than classical clustering algorithm with different dis-tributions and different feature contributions.