Enhancing one-class classification performance through variable selection: A review based on advanced literature search approaches

Variable selection is a key step in improving One-Class Classification (OCC), especially when applied to high-dimensional datasets common in chemometrics and anomaly detection tasks. This systematic literature review explores how different strategies—filter, wrapper, embedded, and hybrid methods—have been employed to enhance OCC models' accuracy, interpretability, and robustness. A comprehensive search was conducted using Scopus, complemented by AI-powered tools such as Elicit and Litmaps, and visual analytics platforms including VOSviewer and Bibliometrix. The review highlights methodological trends across both chemometric and machine learning domains, revealing a predominance of embedded approaches and a growing interest in hybrid strategies. Embedded methods, particularly LASSO, Elastic Net, and autoencoder-based architectures, were favored for their scalability and model integration. Approximately 69 % of the reviewed studies adopted a rigorous OCC approach—relying solely on target class data—demonstrating a preference for bias-resistant modeling. Additionally, bibliometric analysis revealed a disciplinary division, with chemometric studies emphasizing analytical applications and model interpretability, while computer science-driven studies prioritized automation and scalability. The findings emphasize the need for flexible, domain-adapted variable selection pipelines capable of handling class imbalance and high dimensionality. This work also introduces a reproducible framework combining traditional and AI-assisted literature review tools to support future systematic analyses. The review concludes by identifying emerging trends and suggesting future research directions in OCC and variable selection, with a focus on hybrid modeling, domain adaptability, and performance benchmarking across application fields.