DDSIMCA with LOVE

Variable selection (VS) is a critical step in improving the performance of chemometric<br/>models. While well-established VS methods exist for regression and discrimination, their applicability to one-class classifiers (OCC) remains limited. This study introduces LOVE (Leave One Variable Excluded), a novel VS method specifically designed for OCC, addressing the existing gap in chemometric modeling.<br/><br/>LOVE is a wrapper-based VS approach that integrates interactivity, allowing for iterative refinement of selected variables based on multiple performance indicators. Unlike automated methods, LOVE enables analysts to examine and adjust selection decisions at each step, fostering a more informed and flexible modeling process. The algorithm operates in two modes: discrete (individual variable selection) and interval (grouped variable selection for collinear data, e.g., spectra). The procedure iteratively removes variables or intervals while evaluating performance indicators such as sensitivity, specificity, selectivity and efficiency, ensuring robust classifier optimization.<br/><br/>LOVE was evaluated using several case studies across different analytical platforms, demonstrating its ability to:<br/><br/>1. Enhance classifier performance by selecting optimal subsets of variables.<br/>2. Prevent overfitting and improve model stability.<br/>3. Handle outliers without direct exclusion.<br/>4. Provide deeper insight into influential variables affecting classification decisions.<br/><br/>The interactive nature of LOVE distinguishes it from existing VS methods, offering greater transparency and adaptability in OCC. The ability to adjust performance indicators and explore variable interactions in real-time makes it particularly suitable for chemometric applications. Furthermore, LOVE has been implemented as a freely available web-based tool for DD-SIMCA, facilitating its adoption in diverse analytical scenario