Machine Learning Analysis of Livestock Depredation and Carnivore Coexistence

Abstract

One of the most common and long-standing sources of human–wildlife conflict in pastoral settings is the depredation of livestock by large carnivores, which poses risks to both livelihoods and the long-term survival of threatened predator populations. The study presents a conceptual, data-driven approach to the analysis of livestock depredation and coexistence conditions based on ecological, household, and community-level data gathered from livestock-loss records, livestock household interviews, field observations, compensation claims, and wildlife occurrence data in a mixed pastoral–carnivore landscape. The study does not provide a technical description of the model architecture but instead sets the machine learning work in the context of a tool that identifies the complex, non-linear patterns of risk that are often missed by traditional statistical summaries, such as the influence of the combination of household attitude, seasonal grazing patterns, herd management practices and proximity to habitat on the risk of depredation. The findings show that the risk of depredation is greatest in communities close to grazing areas at the forest edge, that herds with poor enclosure or community management are particularly vulnerable, and that community tolerance is linked to satisfaction with compensation and previous experience of loss. The results indicate that the problem of livestock depredation is not exclusively ecological; it must be approached with an integrated perspective, taking into account the ecological context, husbandry practices and social perception. The paper concludes that a coexistence-oriented approach is the most viable path forward moving towards a reduction of conflict without compromising carnivore conservation in shared landscapes, to be achieved through strengthened enclosures, supervised grazing, locally appropriate deterrent measures, transparent compensation, and active participation of the community