Optimization of K-means Clustering by Feature Weight Learning

The performance of K-means clustering algorithm depends on the selection of distance metrics. The Euclidean distance is usually chosen as the similarity measure in the conventional K-means clustering algorithm, which usually relates to all attributes. When feature weight parameters are introduced to the distance formula, the performance will depend on the weight values and accordingly can be improved by adjusting weight values. Since K-means algorithm is iterative, it is difficult to optimize clustering results by giving weight values directly. An indirect learning feature weight algorithm is introduced to improve the clustering result. Mathematically it corresponds to a linear transformation for a set of points in the Euclidean space. The numerical experiments prove the validity of this algorithm.