RLAC: Random Line Approximation Clustering

Abstract

We explore how Random Projections can be used as an Approximate method for Projection Pursuit Clustering in high dimensional data. Traditional data transformations such as PCA for dimensionality reduction have been shown to be beneficial in clustering. However, their objective is not always relevant to the cluster structure producing undesirable results. On the other hand, Projection Pursuit methods present promising results in finding different "interesting"directions while being easily modified, though they came with high computational costs. In an attempt to provide a lightweight and simplified approach for Projection Pursuit clustering, we designed and implemented the Random Line Approximation Clustering (RLAC), a hierarchical divisive clustering algorithm that incorporates attributes from the Random Projection method. © 2021 IEEE.