Rapid prediction of essential oils major components by Vis/NIRS models using compositional methods

Abstract

Essential oils (EOs) have often a major compound, which content is important for the industry. Thus, rapid techniques suitable to analyze the EOs composition are of great interest. Near infrared spectroscopy (NIRS) has proven suitable solutions for many similar targets. NIRS depends on multivariate correlation methods, which were designed for the real space, with values comprised between - ∞ and + ∞. However, essential oils compounds are compositional data (CoDa), restricted to a simplex space. These are co-dependent data, therefore their multivariate correlation with the spectral variables needs specific methods. This study evaluates predictive models of the contents of cinnamaldehyde, eugenol methyl-ether and thymol, respectively in EOs of cinnamon, clove and thyme, built by using CoDa method. This last was the isometric log-ratio transformation (ILR). Besides, the calibrations set without CoDa methods, were assessed. The results showed clearly the best prediction in external validation exercises was by using ILR. This predictive exercise pointed out the compositional method provided better model robustness than the calibration without this.

Keywords

• CoDa
• Compositional data
• Essential oil
• NIRS