<![CDATA[Recommendation systems increasingly employ hybrid learning pipelines to address data sparsity, cold-start problems, high-dimensional user-item features, and unstable preference signals. Feature selection is essential because irrelevant and redundant variables reduce prediction accuracy, increase computational cost, and limit model interpretability. Although Principal Component Analysis (PCA) and Mutual Information (MI) are widely applied for dimensionality reduction and feature relevance evaluation, their combined role in recommendation systems remains underexplored. This systematic review synthesized evidence on hybrid feature-selection approaches and the potential integration of PCA and MI in recommendation pipelines following the PRISMA 2020 guidelines. A PRISMA-based screening process identified 552 records, removed 147 duplicates, screened 405 titles and 169 abstracts, assessed 74 full-text articles, and included 30 studies for qualitative synthesis. Three dominant themes emerged: hybrid recommender models integrating collaborative and content-based filtering, hybrid feature-selection frameworks combining filter and wrapper methods, and PCA-MI pipelines for high-dimensional classification. Although direct PCA-MI applications in recommendation systems remain limited, existing evidence supports a two-stage framework in which PCA extracts stable latent representations and MI ranks features based on predictive relevance. Integrating PCA, MI, and wrapper or metaheuristic optimization offers a promising strategy to improve recommendation accuracy, scalability, and explainability across diverse application domains.]]>
Copyrights © 2026
