Performance Evaluation of AI-Enhanced Unsupervised CART Models for High-Dimensional Data Classification

Abstract

This study evaluates the performance of AI-enhanced unsupervised Classification and Regression Tree (CART) models for high-dimensional data classification, addressing the growing need for interpretable and scalable unsupervised learning techniques in domains where labeled data are limited or unavailable. Traditional clustering and dimensionality reduction methods often struggle with nonlinear relationships, noise, and complex feature interactions, motivating the integration of advanced artificial intelligence mechanisms into CART frameworks. The research investigates how deep learning–based feature representation, hybrid clustering-tree architectures, reinforcement learning–driven decision policies, and evolutionary optimization contribute to improved clustering accuracy, boundary formation, and structural adaptability. Using multiple benchmark datasets representing healthcare, cybersecurity, remote sensing, and text analytics, the study compares enhanced CART models with conventional unsupervised algorithms based on metrics such as silhouette coefficient, Davies–Bouldin index, cluster purity, cohesion, and interpretability. Experimental results demonstrate that AI-augmented CART variants significantly outperform standalone methods by producing more coherent clusters, reducing sensitivity to noise, and offering transparent rule-based explanations