Variable selection for one class classifiers. Introduction of LOVE
Alexey L. Pomerantsev, Sergey Kucheryavskiy, Oxana Ye. Rodionova · Analytica Chimica Acta · 2025
BACKGROUND: Variable selection methods for regression and discrimination are well developed. However, the known methods do not adequately address the problem in the case of one class classifiers. This study aims to fill this gap. RESULTS: A new variable selection method, LOVE (Leave One Variable Excluded), created specifically for one class classifiers is proposed. LOVE belongs to the wrapper family and is interactive, in contrast to most known methods, which are automatic. The four cases considered demonstrate that the method performs well in various scenarios. SIGNIFICANCE: It is shown that LOVE can: (1) enhance classifier performance; (2) prevent overfitting and improve model stability; (3) fix outliers without deletion; and (4) provide a better understanding of what factors influence the decision.