Super-klust: Another Way of Piecewise Linear Classification
This work offers an incremental simplification of an existing piecewise-linear classification method for machine learning researchers.
This paper introduces Super-klust, a piecewise-linear classification algorithm based on Voronoi tessellations. It simplifies the Super-k algorithm by replacing its voxelization and expectation-maximization stages with a distance-based clustering algorithm, such as k-means. Experimental results indicate that Super-klust achieves similar performance to Super-k.
With our previous study, the Super-k algorithm, we have introduced a novel way of piecewise-linear classification. While working on the Super-k algorithm, we have found that there is a similar, and simpler way to explain for obtaining a piecewise-linear classifier based on Voronoi tessellations. Replacing the multidimensional voxelization and expectation-maximization stages of the algorithm with a distance-based clustering algorithm, preferably k-means, works as well as the prior approach. Since we are replacing the voxelization with the clustering, we have found it meaningful to name the modified algorithm, with respect to Super-k, as Supervised k Clusters or in short Super-klust. Similar to the Super-k algorithm, the Super-klust algorithm covers data with a labeled Voronoi tessellation, and uses resulting tessellation for classification. According to the experimental results, the Super-klust algorithm has similar performance characteristics with the Super-k algorithm.