A weighted k-mean clustering algorithm based on singular values with offset clustering centers

Abstract

The K-means algorithm is widely used for dataset clustering, but it does not consider the importance of each attribute dimension when dealing with feature attributes and clustering center selection, but rather treats all attributes as having equal importance. In order to solve this problem, this paper proposes a weighted k-mean clustering algorithm (SVW-KMeans) based on singular values with offset clustering centers. The algorithm calculates the weight information of the data points through singular value decomposition to focus on the most significant and most different features, joining the weight calculation to optimize the objective function, and at the same time, the weighted arithmetic mean of the individuals is used as the clustering center, and the clustering center is shifted towards the high importance so as to take into full consideration of the importance of the different features in the clustering process. The experimental results show that the SVW-KMeans algorithm outperforms other algorithms in clustering on synthetic and real datasets, which verifies that the SVW-KMeans algorithm outperforms other mainstream clustering algorithms in terms of clustering quality and stability.