We present the utilization of a reduction property found in the penalty-based formulation of pseudo-Boolean polynomials as a means to reduce dimensionality in cluster analysis procedures. Through our experiments, we demonstrate that multidimensional datasets, such as the 4-dimensional Iris Flower dataset, can be condensed into 2-dimensional space, while the 30-dimensional Wisconsin Diagnostic Breast Cancer (WDBC) dataset can be condensed into 3-dimensional space. By identifying lines or planes that exist between the condensed samples, we are able to extract clusters in an unbiased and linear manner, achieving competitive accuracies, reproducibility, and clear interpretation.
Cluster Analysis for Dimensionality Reduction through pseudo-Boolean Polynomials (arXiv:2308.15553v1 [cs.IR])
