Reproduce “An Approach to Variable Clustering: K-means in Transposed Data and its Relationship with Principal Component Analysis” on OpenResearch

About this paper

Victor Saquicela, Kenneth Palacio-Baus, Mario Chifla

This research introduces a methodological framework that integrates K-means variable clustering on transposed data with Principal Component Analysis, quantitatively assessing how groups of variables contribute to each principal component. The approach established clear relationships in datasets with simple structures and provided complementary insights into variable group influences in more complex data.

Key insights

• The degree of correspondence between K-means variable clusters and high-loading PCA variables is dependent on the inherent variability structure of the data.
• Datasets with simple, well-defined structures exhibit a strong, interpretable one-to-one mapping between variable clusters and specific principal components.
• Even in complex datasets where one-to-one mapping is 'fuzzy', the method provides valuable complementary insights by revealing the joint behavior and relative importance of variable groups across principal dimensions.

Summary by alphaXiv.