Supervised machine learning aided behavior classification in pigeons.

Area of Science:

• Ethology
• Computational Biology
• Bioinformatics

Background:

• Manual behavioral observations are time-consuming, labor-intensive, and prone to subjectivity, hindering reproducibility in ecological and neuroscientific research.
• Existing automated behavioral analysis tools primarily focus on mammals, limiting applications for other vertebrate groups like birds.
• Extending automated analysis to birds is crucial for species-specific knowledge and understanding evolutionary behavior patterns.

Purpose of the Study:

• To present an open-source software package for automated bird behavior classification.
• To enable the analysis of pose-estimation data for supervised machine learning models.
• To facilitate the extension of automated behavioral analysis to diverse avian species.

Main Methods:

• Utilized pose-estimation data generated by deep-learning tools (e.g., DeepLabCut).
• Developed supervised machine learning classifiers for pigeon behaviors.
• Trained various machine learning and deep learning architectures using multivariate time series data.

Main Results:

• Achieved a high F1 score of 0.874 for classifying seven distinct pigeon behaviors.
• Demonstrated the effectiveness of machine and deep learning architectures on time series pose-estimation data.
• Introduced an algorithm for tuning classifier bias towards precision or recall.

Conclusions:

• The developed open-source software provides a foundation for automated bird behavior classification.
• This approach can be broadened to support a wider range of bird species.
• The tool offers flexibility in tailoring behavioral classifiers to specific research needs.