Development of a Machine Learning Based Framework for Investigating Environmental Influences on Avian Sperm Morphology

ABSTRACT
This project addresses a critical gap in understanding climate-driven reproductive adaptation by investigating the influence of bioclimatic variables on avian sperm morphology across temperate and tropical regions. Using a dataset of 2,704 avian specimens with morphometric and geoclimatic data, we developed an interpretable machine learning framework to uncover adaptation signatures beyond the limits of traditional morphometric analysis. The pipeline integrates Random Forest regression for predictive modeling and SHAP (SHapley Additive exPlanations) for feature interpretation. Through region-based comparative analysis and engineered morphological ratios, the study reveals how environmental gradients, particularly precipitation seasonality and annual precipitation, shape variation in sperm traits. Our best models achieved R² scores ranging from 0.13 to 0.54, with proportion-based traits (e.g., Flagellum_to_Head, Midpiece_to_TotLength) and total length metrics showing stronger predictive relationships. Despite some unexplained variance, SHAP interpretation revealed distinct climate-trait relationships across regions, offering empirical evidence of climate-linked adaptation. Implemented in Python using Pandas, Seaborn, Scikit-Learn (For Random Forest), and SHAP, this project contributes a scalable computational approach to avian evolutionary analysis and lays the groundwork for future climate-informed reproductive and conservation studies.