High-dimensional pattern extraction has become a critical challenge in data-mining applications within
Computer Science and Engineering, especially in cybersecurity and anomaly-driven environments.
This study presents a concise comparative analysis of classical Machine Learning (ML) models and
AI-enhanced Deep Learning (DL) embedding learners for robust pattern discovery on complex highdimensional
feature spaces. Experiments were conducted on two established datasets—NSL-KDD (41
features, 25 attack classes) for intrusion mining and the Kaggle Credit Card Fraud dataset (30 PCA
features, 284,807 records) for scalability evaluation. ML models including SVM-RBF, Random Forest
(200 trees), XGBoost, and LightGBM were benchmarked against DL models—ANN-MLP and 1D-CNN
embedding learners—for pattern-fidelity, runtime efficiency, and memory footprint. Results indicate that
tree-based learners achieve superior accuracy on engineered feature spaces, while deep embedding
models generate richer, compressed latent patterns that enhance mining stability. The evaluation
covered accuracy, precision, recall, F1-score, training time, runtime feasibility, and memory usage,
supported through multiple comparative visualizations. The findings demonstrate that AI enhances data
mining by improving latent pattern-visibility, convergence stability, clustering density, scalability, and
resource-efficiency, providing meaningful insights for real-world high-dimensional mining deployments
in CSE domains.
