Non-Invasive Body Mass Estimation in the Ladybird Spider, Eresus kollari, Using Multi-View Morphometrics and Machine Learning

Abstract

Body mass is widely used as an indicator of physiological condition and ecological fitness in spiders. However, directly measuring the mass of small, live individuals remains technically challenging. In this study, we evaluated a non-invasive approach for predicting body mass in Eresus kollari using machine learning-based (ML) models derived from morphometric traits. Using 432 records of body size measurement data, we trained models for four view types (dorsal, ventral, oblique, and lateral) and compared their performance with equation-based (Eq) models. For each view type, multiple regression models were evaluated, and tree-based ensemble models were selected across all view types: Extra Trees Regressor for the dorsal, ventral, and lateral views, and Random Forest Regressor for the oblique view. The ML models showed lower prediction error and higher predictive consistency than the Eq models. Feature importance analysis revealed that body length-related features were among the most influential predictors in models incorporating this trait. The oblique view models, which do not include body length as a predictor, showed competitive performance but were less effective at detecting the feeding-related mass increase in an illustrative demonstration. Although the suitability of view-specific models may vary across species and developmental stages depending on the primary mode of body size change, our results suggest that ML-based approaches offer a promising, practical approach for non-invasive estimation of spider body mass in behavioral and ecological research.