Enhancing the classification of isolated theropod teeth using machine learning: a comparative study.

Area of Science:

• Paleontology
• Computer Science
• Machine Learning

Background:

• Classifying fossilized theropod teeth using morphometric variables is challenging due to time constraints and intra-class variability.
• Machine learning (ML) offers a potential solution for automating the taxonomic identification of isolated teeth.

Purpose of the Study:

• To compare the performance of six ML techniques for classifying theropod teeth at genus and higher taxonomic levels.
• To evaluate the impact of data standardization and oversampling techniques on ML model performance with imbalanced datasets.
• To differentiate teeth based on their position in the tooth row.

Main Methods:

• Utilized datasets with morphometric features from isolated theropod teeth.
• Applied and compared six distinct ML classification techniques.
• Investigated the effects of various standardization and oversampling methods on imbalanced datasets.

Main Results:

• Performance varied significantly among ML models, with some demonstrating higher sensitivity to class imbalance.
• Specific standardization and oversampling techniques improved classification accuracy for certain models.
• The study identified optimal ML approaches for theropod tooth classification.

Conclusions:

• Machine learning provides a viable method for automating theropod tooth classification.
• Addressing class imbalance through appropriate data preprocessing is crucial for reliable ML model performance.
• Publicly available models and methods facilitate future research and cross-study comparisons in paleontology.