Variable Selection for Skewed Clustering and Classification

Abstract

As datasets from virtually all fields of endeavour continue to grow in size and complexity, the curse of dimensionality cannot be overlooked. Researchers in model-based clustering have recognized the need for effective dimension reduction techniques; as a result, many such algorithms exist to date. These algorithms, however, are often specific to Gaussian clustering problems and break down in the presence of skewness. We present a novel skewed variable selection algorithm that utilizes the Manly transformation mixture model to select variables based on their ability to separate clusters. We compare our approach with other asymmetric and normal variable selection methods using simulated and real-world datasets. We find that the proposed algorithm is suitable for dimension reduction in the presence of skewness.