Development of an amazonian oils database for statistical analysis and machine learning modeling

This article aims to develop an Amazonian oils database to support statistical analysis and machine learning modeling. Data on fatty acid composition and physicochemical properties of 65 vegetable oils and fats, including 17 Amazonian species, were compiled. From individual fatty acid profiles, diversity indices, correlation matrices between composition and physicochemical properties, principal component analysis (PCA) and hierarchical cluster analysis (HCA) were obtained. These approaches enabled the description of the lipid space covered by the evaluated oils and the positioning of Amazonian oils relative to conventional and exotic vegetable matrices. Correlations between SFA, MUFA and PUFA levels and parameters such as melting point, iodine value and oxidative stability were consistent with established principles of lipid chemistry. The analysis restricted to the 17 Amazonian oils revealed distinct chemotypes, including a lauric–myristic group (murumuru, ucuúba, babassu), an oleic–linoleic group (açaí, bacaba, andiroba, buriti, patauá, tucumã, pracaxi, bacuri and cupuaçu) and the singular profile of copaiba. Additionally, a supervised XGBoost (XGBClassifier) model was used to discriminate Amazonian and non‑Amazonian oils based solely on individual fatty acids. SHAP‑based interpretation indicated that gadoleic, caprylic, linoleic, oleic, lauric and myristic acids play a prominent role in characterizing the Amazonian set. Overall, the database, combined with statistical and machine learning tools in Python/Jupyter, represents a promising approach to map the compositional space of vegetable oils and to identify Amazonian lipid signatures with technological potential.