Data-driven modelling of Aquilaria essential oils via dual GC profiling and multicollinearity diagnostics

Abstract

Aquilaria-derived essential oils are chemically diverse and hold significant value in pharmaceuticals, fragrances, and traditional medicine. However, the complexity of their chemical composition presents challenges in statistical modelling, particularly due to multicollinearity among biosynthetically related compounds. This study investigates the extent of multicollinearity in Aquilaria essential oil data using multiple linear regression (MLR) and variance inflation factor (VIF) analysis. A regression model was constructed using three compounds, ? -guaiene, 10-epi ? -eudesmol, and ? -eudesmol, across 360 samples, with VIF and collinearity diagnostics applied to assess model validity. The model explained 93% of the variance in species classification, which is substantially higher than values typically reported in earlier chemometric studies of Aquilaria oils. This demonstrates that even a limited number of carefully selected compounds, when supported by diagnostic safeguards, can achieve strong classification accuracy. These findings emphasize the importance of applying multicollinearity diagnostics to improve the interpretability and reliability of chemometric analyses. The study contributes a robust analytical framework for future research and practical applications in species authentication, essential oil quality control, and conservation of Aquilaria resources.