Predicting Elephant Crop-Raiding with Machine Learning to Inform Conservation Planning: A Case Study Comparing the Performance and Tradeoffs of GeoAI Algorithms in Northwestern Zimbabwe
Abstract
Human–elephant conflict (HEC) is a persistent socio-environmental challenge arising from the intersection of wildlife ecology, land use change, and development. African savanna elephants (Loxodonta africana) come into conflict with humans when space and resource needs overlap. Patterns of HEC are complex and the relationships between drivers and conflict are not always linear. In northwestern Zimbabwe, south of the city of Victoria Falls, conflict arises as the rainy season comes to a close and crops mature. In this study, we demonstrate how Geospatial Artificial Intelligence (GeoAI) can be integrated into environmental decision-making to model and precisely mitigate conflict. We combined geospatial datasets with machine learning (ML) algorithms to predict crop-raiding occurrence. Using over 700 crop-raiding conflict locations, we identified anthropogenic and natural factors as model predictors. We compared random forest, support vector machine, and neural network approaches within a geospatial pipeline and evaluated model performance. Results from different ML algorithms reveal nonlinear and moderately complex patterns within our dataset. The likelihood of conflict increases with closer proximity to protected areas and further from anthropogenic activity. When predicting conflict, multi-layer perceptron neural networks achieved the highest accuracy (95.5%), outperforming support vector machines (76.3%), random forests (81.5%), and deep neural networks (69.0%). While it does not provide insight into individual predictor contributions, the multi-layer perceptron neural network model outperformed all others on testing data, suggesting it will scale to other regions. Beyond prediction, our work emphasizes the promise of GeoAI as a decision-support tool. As agricultural area expands and the spatial and resource needs of humans and elephants increasingly overlap, the accurate and targeted conflict prediction provided by GeoAI models can facilitate mitigation methods that are locally salient, evidence-based, and cost-effective. Our work provides transferable insights for integrating GeoAI into environmental management to address human–wildlife conflict and other spatially complex environmental challenges.
Related
- Predicting visual impairment in type 2 diabetes using key performance indicators identified by the Black Widow and Whale Optimization Algorithms: A case study in Saudi Arabia
- Comparing advanced interference suppression algorithms for OPM-MEG
- Panorama: A robust pangenome-based method for predicting and comparing biological systems across species
- MF-TLMLP: A multi-channel fusion-based multi-fidelity transfer learning method for predicting the cutting performance of conical picks
- CaliPPer: quantifying, predicting and improving AI model performance for binding prediction
BibTeX
@misc{markham2026predicting,
title = {{Predicting Elephant Crop-Raiding with Machine Learning to Inform Conservation Planning: A Case Study Comparing the Performance and Tradeoffs of GeoAI Algorithms in Northwestern Zimbabwe}},
author = {Katherine Elizabeth Markham and Sergio Bernardes and Ferrel V. Osborn and Marguerite Madden and Malvern Karidozo and William R. Langbauer and Angela Yao and Andrew Grundstein and Roger Parry},
howpublished = {EarthArXiv},
year = {2026},
month = sep,
doi = {10.31223/x5gr57},
url = {https://eartharxiv.org/repository/view/15144/},
note = {CC BY-NC 4.0},
}
RIS
TY - UNPB
TI - Predicting Elephant Crop-Raiding with Machine Learning to Inform Conservation Planning: A Case Study Comparing the Performance and Tradeoffs of GeoAI Algorithms in Northwestern Zimbabwe
AU - Katherine Elizabeth Markham
AU - Sergio Bernardes
AU - Ferrel V. Osborn
AU - Marguerite Madden
AU - Malvern Karidozo
AU - William R. Langbauer
AU - Angela Yao
AU - Andrew Grundstein
AU - Roger Parry
PY - 2026
DA - 2026/09/23
PB - EarthArXiv
DO - 10.31223/x5gr57
UR - https://eartharxiv.org/repository/view/15144/
AB - Human–elephant conflict (HEC) is a persistent socio-environmental challenge arising from the intersection of wildlife ecology, land use change, and development. African savanna elephants (Loxodonta africana) come into conflict with humans when space and resource needs overlap. Patterns of HEC are complex and the relationships between drivers and conflict are not always linear. In northwestern Zimbabwe, south of the city of Victoria Falls, conflict arises as the rainy season comes to a close and crops mature. In this study, we demonstrate how Geospatial Artificial Intelligence (GeoAI) can be integrated into environmental decision-making to model and precisely mitigate conflict. We combined geospatial datasets with machine learning (ML) algorithms to predict crop-raiding occurrence. Using over 700 crop-raiding conflict locations, we identified anthropogenic and natural factors as model predictors. We compared random forest, support vector machine, and neural network approaches within a geospatial pipeline and evaluated model performance. Results from different ML algorithms reveal nonlinear and moderately complex patterns within our dataset. The likelihood of conflict increases with closer proximity to protected areas and further from anthropogenic activity. When predicting conflict, multi-layer perceptron neural networks achieved the highest accuracy (95.5%), outperforming support vector machines (76.3%), random forests (81.5%), and deep neural networks (69.0%). While it does not provide insight into individual predictor contributions, the multi-layer perceptron neural network model outperformed all others on testing data, suggesting it will scale to other regions. Beyond prediction, our work emphasizes the promise of GeoAI as a decision-support tool. As agricultural area expands and the spatial and resource needs of humans and elephants increasingly overlap, the accurate and targeted conflict prediction provided by GeoAI models can facilitate mitigation methods that are locally salient, evidence-based, and cost-effective. Our work provides transferable insights for integrating GeoAI into environmental management to address human–wildlife conflict and other spatially complex environmental challenges.
ER -