Machine Learning Algorithms for Automatic Classification of Marmoset Vocalizations.

Area of Science:

• Primate vocalization analysis
• Bioacoustics
• Machine learning in animal behavior

Background:

• Acoustic monitoring of captive primates is valuable for research.
• Effective automatic classification of vocalizations requires algorithms trainable on small datasets.
• Current methods may lack robustness for small-scale training.

Purpose of the Study:

• To evaluate seven classification algorithms for primate vocalization type.
• To identify a robust classifier suitable for small datasets.
• To facilitate practical acoustic monitoring of captive primate colonies.

Main Methods:

• Comparative analysis of seven distinct classification algorithms.
• Training and testing algorithms on primate vocalization datasets.
• Performance evaluation using accuracy and F1-score metrics.

Main Results:

• The Optimum Path Forest classifier demonstrated high performance.
• Achieved accuracy exceeding 0.83 and F1-score above 0.84.
• Successfully trained on small datasets, indicating robustness.

Conclusions:

• Optimum Path Forest is a reliable tool for automatic primate vocalization classification.
• The developed method supports acoustic monitoring of captive primate populations.
• Publicly releasing the dataset and algorithms promotes further research.