Reproduce “Machine learning approaches to explore important features behind bird flight modes” on OpenResearch

About this paper

Yukino Kawai, Tatsuya Hisada, Kozue Shiomi, Momoko Hayamizu

Researchers systematically applied modern machine learning interpretability methods to quantify the importance of morphological, physiological, and ecological features in determining avian flight modes. Their analysis of 635 migratory bird species showed that incubation period and wing length are the primary determinants of flight classification, with the LightGBM model achieving 97.17% accuracy.

Key insights

• Incubation period and wing length were consistently identified as the most important features in determining avian flight mode across different machine learning interpretability methods.
• Life history traits, such as developmental periods, can be as important as morphological features in shaping avian flight strategies.
• Machine learning-derived feature importance values can capture and reveal biologically meaningful relationships and influence species clustering patterns, even without explicit phylogenetic information.

Summary by alphaXiv.