A New Algorithm for Robust Affine-Invariant Clustering

Abstract
Cluster analysis is an unsupervised machine learning technique commonly employed to partition a dataset into distinct categories referred to as clusters. The k-means algorithm is a prominent distance-based clustering method. Despite its overwhelming popularity, the algorithm is neither invariant under non-singular linear transformations nor robust, i.e., can be unduly influenced by outliers. To address these deficiencies, we adopt an alternative clustering procedure based on minimizing a “trimmed” variant of the negative log-likelihood function and develop a novel “concentration step” (C-step) that can iteratively reduce the objective function. A simulation study over multiple synthetic scenarios and a real-world example are analyzed to assess the performance of our algorithm. Compared to k-means, empirical studies indicate competitiveness and oftentimes superiority of our algorithm. This is a joint work with Michael Pokojovy (The University of Texas at El Paso). 

About the Speaker

Andrews T. Anum is currently a final year PhD candidate in the Computational Science Program at The University of Texas at El Paso (UTEP). His research revolves around developing, implementing and applying new statistical and machine learning methodologies for multivariate, nonparametric and robust statistics to provide solutions and enhance decision making in engineering, healthcare, and finance, to name just a few examples.