Machine Learning for identification and classification of Foraminifera: testing on monothalamids

Here we propose an AI-based approach using Machine Learning (ML) to assist species identification and reduce morphotype redundancy in the study of monothalamous foraminifera. In fact, this group of protists, is often overlooked in taxonomic studies due to their morphological simplicity and diversity. These single-celled organisms with "soft" tests are poorly studied, with only a few species identified, while many morphotypes remain undescribed. Taxonomic research on monothalamids is limited by challenges in identification, lack of fossilization, and the time-intensive nature of the work. This gap may lead to underestimating biodiversity and hinder detecting ecosystem degradation. Despite these challenges, monothalamids play key roles in marine ecosystems, making their diversity crucial for conservation and resource management. With this in mind, we analyzed images from the scientific literature, extracting key morphological traits, such as chamber shape, shell type, composition, and aperture type, through objective human annotation to build a dataset processed by ML algorithms. Clustering techniques, such as K-Means, revealed that basic shape, followed by shell type and composition, were the primary features distinguishing clusters. This approach enabled more objective morphotype classification, improving consistency and reducing human bias. These findings align with recent taxonomic revisions and demonstrate that applying unsupervised ML methods enhances species identification accuracy and streamlines the analysis of high-dimensional datasets.