Machine Learning Prediction of Human–Elephant Conflict Under Land-Use Change

Abstract

Agriculture and settlements expand and elephant habitats become fragmented, human–elephant conflict (HEC) is worsening throughout elephant range landscapes. Reactive management such as fencing, patrolling and compensation problem solve after incidents have occurred but do not always foresee the likelihood of occurrence or in what circumstances. The purpose of this paper is to propose a conceptual framework to predict human–elephant conflict risk given the current land-use change scenario through machine-learning approaches. In this study, a specific algorithmic model for implicating land-use and environmental information to conflict-risk indicators, an assessment stage based on machine learning and classified low, moderate and high-risk zones has not been implemented or tested, but is presented in a strictly conceptual, theory-based framework that synthesizes the ecological, spatial, climatic and human-activity indicators from the wider literature. The framework should be able to identify high-risk landscapes (including fragmented corridors and forest edge settlements), show seasonal conflict patterns related to rainfall and to the cropping cycle and identify major environmental and human drivers of risk. The outputs will be used for early-warning systems and for land-use planning decisions and elephant conservation strategies, to target limited mitigation resources to the areas and times that are most critical. The paper also addresses data issues, ethical problems and field validation issues that need to be overcome to make such a framework operational in a responsible manner. Overall, the study offers a coherent conceptual link between land-use science, conflict ecology and machine-learning, on which future data-driven, empirically validated conflict-prediction systems can be built.