Comparative analysis of the influence of temporal, seasonal and behavioral factors using XGBoost for predicting traveled distances

Cattle farming are essentially based on the exploitation of natural pastures. Determining the distances traveled by animals in extensive livestock farming remains a major challenge in the West African sub-region. The objective of this study was to develop a robust predictive model of the daily distance traveled by animals in pastoral systems, exploiting the capabilities of the XGBoost algorithm. This study evaluated pastoral cattle movements in two municipalities in different agroecological zones of Burkina Faso. Tracking carried out with mini-GPS devices made it possible to collect positioning data on the movement of grazing animals over three distinct seasons. The daily distance traveled was modeled with an XGBoost algorithm whose performance was validated by cross-validation. Post-modeling analyses then made it possible to interpret the model and identify factors influencing herd behavior. Performance was good, with a coefficient of determination R of 0.92 and an average error of only 245 meters. Variable importance analysis revealed a dominance of grazing time (69.4%) and travel speed (22.3%) in predicting distances traveled. The XGBoost method offers a robust alternative to traditional statistical models. This approach opens up concrete prospects for the development of decision-support tools that enable livestock farmers and managers to more effectively plan rangeland use, reduce environmental impact, and improve animal welfare.