Machine Learning for Anti-Poaching: Decision Tree Applications on the Savannah
(2025)Department of Automatic Control
- Abstract
- The use of technology, specifically machine learning, to address complex issues in developing regions has become increasingly prominent. In the context of wildlife conservation, combating poaching has emerged as a significant application of such technology.
This research project aims to develop machine-learning-based low-cost antipoaching solutions. It presents three case studies, each exploring a unique sensor modality and use-case on the savannah, while maintaining a common framework of machine learning algorithms.
1. The first use case is the detection of gunshots using microphone-based systems.
2. The second use case is the detection of elephant footsteps using geophonebased systems.
3. The third use case is the detection of... (More) - The use of technology, specifically machine learning, to address complex issues in developing regions has become increasingly prominent. In the context of wildlife conservation, combating poaching has emerged as a significant application of such technology.
This research project aims to develop machine-learning-based low-cost antipoaching solutions. It presents three case studies, each exploring a unique sensor modality and use-case on the savannah, while maintaining a common framework of machine learning algorithms.
1. The first use case is the detection of gunshots using microphone-based systems.
2. The second use case is the detection of elephant footsteps using geophonebased systems.
3. The third use case is the detection of perimeter fence intrusions using accelerometer-based systems.
Each sensor modality is accompanied by a decision-tree algorithm that aims to optimize the Probability of Detection (POD)-to-False Positive Rate (FPR) ratio. Decision trees are computationally low-cost and intuitive machine learning models that are suitable for systems with low power and memory. The data used in this thesis was collected by the authors at a wildlife sanctuary in South Africa.
The project established high-performing baselines that achieved favorable POD to FPR ratios across two out of three cases. Additionally, the project provided the complete dataset and all machine learning results, including the selected features, the most effective feature selection methods, and an extensive analysis informed by local knowledge from interviewed rangers. This comprehensive approach aids the future professional development of a practically useful system tailored for the savannah environment. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9186769
- author
- Popov Wirén, Jakob Olof and Nordenram, Kasper
- supervisor
- organization
- year
- 2025
- type
- H3 - Professional qualifications (4 Years - )
- subject
- report number
- TFRT-6269
- other publication id
- 0280-5316
- language
- English
- id
- 9186769
- date added to LUP
- 2025-05-09 08:17:25
- date last changed
- 2025-05-09 08:17:25
@misc{9186769,
abstract = {{The use of technology, specifically machine learning, to address complex issues in developing regions has become increasingly prominent. In the context of wildlife conservation, combating poaching has emerged as a significant application of such technology.
This research project aims to develop machine-learning-based low-cost antipoaching solutions. It presents three case studies, each exploring a unique sensor modality and use-case on the savannah, while maintaining a common framework of machine learning algorithms.
1. The first use case is the detection of gunshots using microphone-based systems.
2. The second use case is the detection of elephant footsteps using geophonebased systems.
3. The third use case is the detection of perimeter fence intrusions using accelerometer-based systems.
Each sensor modality is accompanied by a decision-tree algorithm that aims to optimize the Probability of Detection (POD)-to-False Positive Rate (FPR) ratio. Decision trees are computationally low-cost and intuitive machine learning models that are suitable for systems with low power and memory. The data used in this thesis was collected by the authors at a wildlife sanctuary in South Africa.
The project established high-performing baselines that achieved favorable POD to FPR ratios across two out of three cases. Additionally, the project provided the complete dataset and all machine learning results, including the selected features, the most effective feature selection methods, and an extensive analysis informed by local knowledge from interviewed rangers. This comprehensive approach aids the future professional development of a practically useful system tailored for the savannah environment.}},
author = {{Popov Wirén, Jakob Olof and Nordenram, Kasper}},
language = {{eng}},
note = {{Student Paper}},
title = {{Machine Learning for Anti-Poaching: Decision Tree Applications on the Savannah}},
year = {{2025}},
}