Using advanced machine learning to evaluate the spatial and temporal patterns of human–bear interactions in Maryland, USA

Abstract

American black bears ( Ursus americanus ) are highly adaptable to human-dominated areas, which can lead to interactions with humans. Managing these interactions is a priority for many wildlife agencies in North America. Climate-induced shortages in natural food can exacerbate human–bear interactions, as bears are driven to seek alternative food sources in human-dominated areas. However, few studies have investigated the impact of climate or weather on the frequency and severity of human–bear interactions. In Maryland, USA, no research has examined the anthropogenic or environmental factors contributing to human–bear interactions, even though reported interactions have consistently increased since the 1990s. Our research addresses these knowledge gaps by leveraging advanced machine learning techniques, specifically eXtreme Gradient Boosting (XGBoost), to analyze 15 anthropogenic and environmental features in relation to human–bear interactions in Maryland from 2020 through 2023. Through our analysis, we identified meaningful anthropogenic and landscape features that aid in predicting interaction risk and highlighted the often overlooked influence of weather. Our findings revealed an increased risk of interactions during and after droughts, which we visualized on maps that identify predicted high-risk areas across the state. Our research demonstrates the utility of integrating advanced machine learning techniques into ecological studies to achieve high predictive accuracy and inform evidence-based management decisions. The insights and visualizations produced from this effort are designed to support wildlife managers as they assess risk, allocate resources, and develop proactive strategies to foster coexistence between humans and bears in a rapidly changing environment.