A novel machine learning model for the prediction of avian haemosporidian infection from morphological data

Avian haemosporidians are a group of widespread protozoan parasites that often impose significant physiological burdens and reduce host fitness; therefore, an increasing number of studies on these parasites have been performed in recent decades, using different identification methods. Traditional microscopy methods require high-quality blood smears and skilled technicians, whereas molecular methods are complex and costly. In this study, we propose a novel, non-invasive predictive method using supervised machine learning (XGBoost) based on host morphological traits to predict infection status. Using a large dataset of bird banding records and haemosporidian infection results from the Xiaolongmen Forest area in Beijing (2009–2024), we evaluated the associations between infection status and multiple morphological features (e.g., beak length, wing length, and body mass). Our model achieved an average prediction accuracy of 75.3% across eight common bird species, with peak accuracies exceeding 84%. Model validation with independent data from 2023 to 2024 confirmed its robustness. These results suggest that morphological traits, when integrated with machine learning approaches, can serve as effective indicators of haemosporidian infection. The trained models and prediction tools are now available at Github, enabling an immediate infection status prediction based on morphological measurements, and offers a cost-efficient, scalable, and noninvasive alternative to conventional diagnostic methods, with broad applicability in large-scale ecological monitoring, disease surveillance, and conservation planning.