Principal Contrast: Evolutionary Algorithm For Clustering

Author : Chi Tim NG, Huipeng Meng, Yan Wu

Abstract :This paper proposes a novel high-dimensional clustering method that employs an evolutionary algorithm to obtain linear combinations and labeling configurations optimizing the Wilk statistic in the multivariate analysis of variance of the lower-dimensional transformed data. In particular, we consider the situations where the between group variation in each attribute is diminishing, while such a small between group variation is simultaneously reflected in a large amount of attributes. The performance of the proposed method is tested via simulation studies and real data analysis of labeled genetic data from small round blue cell tumors, gene expression profiles from cancer cells, and return data from hedge fund managers

Keywords :High-dimensional clustering, evolutionary algorithm, Wilk statistic, gene expression, cancer data, multivariate analysis, dimensionality reduction, financial data analysis.

Conference Name :International Conference on Mathematics, Statistics, Education & Law (ICMSEL-25)

Conference Place Xian, China

Conference Date 26th May 2025