Testing the accuracy of machine learning methods to predict deforestation

Forest plays a crucial role in meeting climate change goals, given its emissions reduction effects through carbon dioxide capture. The study of deforestation becomes significantly relevant since the early prediction of forest under threat could lead to specific policy responses promoting conservation measures. Common deforestation patterns are fish-bone, radial, geometric, and diffuse. This thesis aims to explore the predictive power of machine learning techniques to predict spatial patterns of human activities and compare their accuracy of prediction with a traditional statistical method. Using Monte Carlo simulations, land cover data was generated, mimicking human settlement patterns related to underlying deforestation processes. This work tests how different machine learning methodologies perform, after various experiments with diverse sources of data. The main result indicates that decision tree-based methodologies provide better prediction performance than other methods including elastic net regression. Implications of this work go beyond the conservation literature and could be used in other agricultural and applied economic areas where spatial patterns play a significant role.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Agricultural & Applied Econ

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2021

Author dc:creator

• Flores Caceres, Ivan Andres

Contributors dc:contributor

• Baylis, Kathy
• Michelson, Hope
• Christensen, Peter

• Machine Learning, Deforestation

Statement dc:rights

• Copyright 2020 Ivan Flores Caceres

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/109618

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/109618