Abstract
This study examines the feasibility of automated classification of geometric objects using multivariate statistical analysis of their parameters, without relying on complex nonlinear machine learning techniques. Objects are described by a set of interrelated geometric and physical characteristics, including linear dimensions, areas, volumes, mass, and density, which leads to high feature correlation and complicates their interpretation.To address this task, classical methods of multivariate data analysis are applied: correlation and regression analysis, principal component analysis (PCA), clustering, and linear discriminant analysis (LDA). The study is conducted on a synthetic dataset formed based on realistic parameter ranges derived from actual geometric relationships.It is demonstrated that preliminary dimensionality reduction using PCA can eliminate multicollinearity among features and produce a compact, interpretable representation of the data. Clustering in the principal component space reveals a stable group structure of objects, while the application of linear discriminant analysis ensures high classification quality. The results confirm that interpretable statistical methods remain an effective tool for analyzing and classifying geometric objects and can be used in tasks of preliminary data analysis, support of expert systems, and development of intelligent ICT-based systems.