属性赋权的K-Modes算法优化

传统K-Modes算法的一个主要问题是属性选择问题。K-Modes算法在聚类过程中对每一个属性都同等看待,而在实际应用中,很多数据集仅有几个重要属性对聚类起作用。为了考虑不同属性对聚类的不同影响,将K-Modes聚类算法与属性权重的最优化结合起来,提出一种属性自动赋权的FW-K-Modes算法。该算法不仅可以提高传统K-Modes聚类算法的聚类精度,还能分析各维属性对聚类的贡献程度,实现关键属性的选择。对多个UCI数据集进行了实验,验证了该算法的优良特性。

Optimization of K-Modes Algorithm with Feature Weights

One major problem of the traditional K-Modes algorithm is the selection of features. The K-Modes clustering algorithm treats all features equally in the clustering process. But in practice, there are only a few important features in many data sets. To consider the particular contribution of different attributes, this paper proposes an improved algorithm called FW-K-Modes algorithm, which incorporates the K-Modes clustering algorithm with feature weight optimization. The proposed algorithm can not only improve the clustering precision in comparison with the traditional K-Modes clustering algorithm, but also analyze the important level of each feature in the clustering pro¬cess and implement the selection of key features. The experimental results on several UCI machine learning data sets validate the effectiveness of the proposed algorithm.