Machine learning recovers folk classification of Banisteriopsis caapi from herbarium leaves an ayahuasca liana.

Ayahuasca refers to an entheogenic brew and its main vine, Banisteriopsis caapi, whose high morphological diversity underlies traditional folk classifications. This study evaluated whether machine learning algorithms can recover folk classifications using images of dried leaves. We analyzed 47 vine plants, mainly B. caapi folk types and Diplopterys cabrerana as an outgroup, using adaxial and abaxial leaf image features related to color, shape, texture, and filters. All evaluated algorithms showed statistically similar performance; however, support vector machine (SVM) achieved the highest accuracy, reaching 70% overall and over 90% for the folk types Hybrid and Arara, and the related outgroup species D. cabrerana. Lower accuracy in Cabi and Quebrador reflects morphological overlap. Confusion matrix and similarity network analyses showed only partial agreement with previous ethnobotanical classifications. Focusing on subtle variation within a single species, this study demonstrates that integrating traditional knowledge with machine learning enables automated validation of folk taxonomies.