A review of machine learning applications for identification and classification problems in paleontology
Carolina S. Marques, Elisabete Malafaia, Soraia Pereira, Vanda Faria dos Santos, Emmanuel Dufourq · Ecological Informatics · 2025
Machine learning (ML) has become an increasingly powerful tool for addressing various challenges in paleontology, including fossil identification and taxonomic classification. In recent years, ML techniques – particularly deep learning and computer vision – have been increasingly adopted for tasks such as morphological analysis, paleoecological inference, and data-driven taxonomic revision. This review synthesizes over one hundred studies across paleontological subfields, establishes key approaches and concepts relevant to paleontologists, and provides practical workflows for applying machine learning techniques to paleontological data. Our study emphasizes that ML improves classification accuracy and helps to overcome long-standing challenges in paleontological research, including observer bias, data scarcity, and subjective interpretations. Most researchers use ML methods to overcome these challenges because they allow the detection of subtle, data-driven patterns, accelerating discovery and expanding the scope of quantitative analysis in paleontology. Nevertheless, reproducibility remains a common problem: only 34.3% of studies presented reproducible research, with just 37.0% making their code available and 56.5% providing public access to their data. While most studies compared different methodologies (57.4%), they often did not use any special pre-processing (36.1%), transfer learning (32.4%), or augmentation techniques (31.5%). Over the years, IT IS REve been methodological advances and improvements in paleontology, demonstrating a shift towards reproducible, scalable, and objective analytical frameworks. This guide aims to support paleontologists in understanding these advances and implementing ML techniques in their own research.