The advantages of k-visibility: A comparative analysis of several time series clustering algorithms

Abstract

This paper outlined the advantages of the k-visibility algorithm proposed in [1,2] compared to traditional time series clustering algorithms, highlighting enhanced computational efficiency and comparable clustering quality. This method leveraged visibility graphs, transforming time series into graph structures where data points were represented as nodes, and edges are established based on visibility criteria. It employed the traditional k-means clustering method to cluster the time series. This approach was particularly efficient for long time series and demonstrated superior performance compared to existing clustering methods. The structural properties of visibility graphs provided a robust foundation for clustering, effectively capturing both local and global patterns within the data. In this paper, we have compared the k-visibility algorithm with 4 algorithms frequently used in time series clustering and compared the results in terms of accuracy and computational time. To validate the results, we have selected 15 datasets from the prestigious UCR (University of California, Riverside) archive in order to make a homogeneous validation. The result of this comparison concluded that k-visibility was always the fastest algorithm and that it was one of the most accurate in matching the clustering proposed by the UCR archive.