Unraveling fungal community biodiversity in table olives through tree-based machine learning classification

In recent years, the number of omics studies aimed at characterizing fungal communities in fermented foods, such as table olives, has steadily increased. Moreover, advances in bioinformatics and artificial intelligence have provided new resources for analysing these datasets. Among these tools, supervised machine learning (ML) models, particularly tree-based algorithms, are especially valuable for their interpretability and their ability to identify underlying patterns. In this study, we evaluated three target variables of table olive samples (processing type, country of origin, and olive cultivar) by implementing three models built with tree-based algorithms, specifically Classification and Regression Trees (CART), Random Forest (RF), and eXtreme Gradient Boosting (XGB). The initial dataset consisted of 872 samples of fungal metataxonomic data obtained from diverse table olive sources. The RF models were the most accurate, achieving an overall accuracy above 75% and a kappa coefficient greater than 0.65. The highest accuracy and robustness were obtained for the model classifying country of origin, with 89% accuracy and a kappa coefficient of 0.86. Furthermore, the use of tree-based models enabled the identification of the fungal genera that contributed most to sample classification. Among the most important genera detected, we noticed Citeromyces , Candida , Pichia , Zygotorulaspora , Taphrina , Wickerhamomyces , Saccharomyces , Starmerella , Aureobasidium , and Dekkera . This approach demonstrates the potential for application of this methodology in the table olive sector and other fermented foods, where the industrial implementation of ML techniques based in omic data could enhance traceability, authenticity, and quality control.