A Machine Learning Approach to Identify Ant Species from the Genus Ectatomma Smith, 1858 (Hymenoptera: Formicidae)

There are some gaps in the taxonomy of certain ant species, making their identification and study more challenging. Regarding some genera, it is an extensive, detailed, and laborious process. This study aims to automate the process of identifying ant species from the genus Ectatomma, utilizing machine learning as a tool, and verifying whether it is possible to make the process more accessible and agile. To this end, the Logistic Classifier, Stochastic Gradient Descent, Random Forest Classifier, k-nearest neighbours, Decision Tree Classifier, Support Vector Classification, and Gaussian Naive Bayes algorithms were applied. The algorithms were implemented in the standard version; therefore, no calibration or changes to internal parameters were made. The data set was divided into 70% for training and 30% for testing. The adaptation of the models to the data set was excellent. All methods applied had positive adaptation in the identification of Formicidae. Only two models did not achieve 100% accuracy, but still maintained accuracy above 80%, which is considered highly positive for ant classification. Four of the six algorithms achieved 100% accuracy, validating the efficacy of these methods for species identification. In terms of Myrmecology, the use of supervised algorithms represents a valuable tool in Taxonomy, especially for species of the genus Ectatomma.