PCA-CLUSTERING: A CLUSTERING AND LABELING METHOD FOR MULTIVARIATE DATA BASED ON THE DEVELOPMENT OF PRINCIPAL COMPONENT ANALYSIS

Abstract

This invention relates to a computational method for multivariate data processing that integrates dimensionality reduction, clustering, and data labeling into a single algorithm architecture based on Principal Component Analysis (PCA). The method is carried out by calculating spatial density based on the eigenvalues ​​of each principal component to produce a division of the data into several clusters. Next, the original feature weights are extracted based on the largest loading vector value in each principal component to obtain a dominant parameter sequence that is used as the basis for labeling each cluster. The clustering results are visualized in two-dimensional or three-dimensional space to facilitate interpretation of the data structure, then validated using reference data that has been labeled by experts in their fields.The validation results are used to evaluate the agreement between the reference labels and the eigenvalue-based clustering results. By integrating all these steps into a single method, this invention simplifies multivariate data processing, reduces the need for separate algorithms, and preserves geometric information and relationships between variables throughout the analysis.