Dataset Creation and Imbalance Mitigation in Big Data - Enhancing Machine Learning Models for Forest Fire Prediction

Aug 3, 2026

Historically, forest fire prediction methods have relied on heuristics, local insights, and basic statistical models, often missing the complex interaction among temperature, humidity, wind speed, and vegetation type. Limited real-time forecasting, combined with highly variable weather patterns linked to climate change, has revealed gaps in traditional approaches across geographically diverse regions such as Canada. This thesis applies machine learning to build an adaptive, data-driven framework using Copernicus reanalysis data and four algorithms for forest fire classification.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages76

Reading Time3 h 12 min

AccessFree to download

1 Introduction

1.1 Motivation

1.2 Objective

1.2.1 Research Question

1.2.2 Contributions

1.3 Thesis Outline

2 Literature Review

2.1 Canada Forest Fire in Early Days

2.1.1 Evolution and Eﬃcacy of Canadian Wildland Fire Information System (CWFIS)

2.2 Forest Fire and Machine Learning (ML)

2.2.1 Data Collection

2.2.2 Imbalance Handling Techniques

2.2.3 ML in Forest Fire Domain

3 Methodology

3.1 Overview

3.2 Notion

3.2.1 Haversine Formula

Dataset Creation and Imbalance Mitigation in Big Data: Enhancing Machine Learning Models for Forest Fire Prediction by Fatemeh Tavakoli A thesis presented to the University of Waterloo in fulﬁllment of the thesis requirement for the degree of Master of Applied Science in Electrical and Computer Engineering Waterloo, Ontario, Canada, 2023 © Fatemeh Tavakoli 2023 Author’s Declaration I hereby declare that I am the sole author of this thesis. This is a true copy of the thesis, including any required ﬁnal revisions, as accepted by my examiners. I understand that my thesis may be made available electronically to the public. Abstract Historically, forest ﬁre prediction methods have leaned on heuristics, local insights, and basic statistical models, often neglecting the complex interplay of variables such as temperature, humidity, wind speed, and vegetation type. The lack of real-time prediction capabilities, paired with unpredictable weather patterns attributed to climate change, underscores the shortcomings of traditional methods, especially in geographically varied regions like Canada. In contrast, machine learning provides the adaptability needed for real-time responses, e↵ectively harnessing updated data and addressing region-speciﬁc forest ﬁre risks. The shift towards machine learning is both a timely and revolutionary approach. This research addresses the urgent need for e↵ective forest ﬁre prediction and management strategies, speciﬁcally in the Canadian context, by harnessing machine learning methodologies. Using Copernicus’s reanalysis data, this study establishes a comprehensive predictive framework employing four cutting-edge machine learning algorithms. Random Forest, XGBoost, LightGBM, and CatBoost. The study features a robust data preprocessing pipeline, class imbalance correction, and rigorous model evaluation measures. Key contributions include the creation of a feature-rich dataset, comprehensive methods for addressing the class imbalance in large scale datasets, and the development of a machine learning framework tailored for forest ﬁre classiﬁcation. The ﬁndings have signiﬁcant implications for data-driven forest management strategies, with the aim of facilitating proactive ﬁre prevention measures on a large scale. One primary challenge encountered was the inherent class imbalance in ﬁre classiﬁcation datasets, with a striking 158:1 ratio between “non-ﬁre” and “ﬁre” events. To address this, the study utilized various re-sampling strategies, encompassing under-sampling, oversampling, and hybrid techniques. Speciﬁc methods employed included NearMiss, SMOTE, and SMOTE-ENN. The NearMiss method with a 0 .09 sampling ratio was found to be particularly e↵ective in addressing this imbalance. When combined with NearMiss version 3 at a 0.09 ratio, the XGBoost model outperformed its peers, showcasing an accuracy of 98.08%, a sensitivity of 86 .06%, and a speciﬁcity of 93 .03% . The ﬁndings indicate that while high recall from NearMiss Version 3 optimized sensitivity, there was sometimes a trade-o↵ with precision. Acknowledgements I would like to express my deepest gratitude to my supervisor, Dr. Sagar Naik, for his unwavering support, insightful critiques, and patient mentorship throughout this research journey. Special thanks go to my committee members, Dr. Ayman El-Hag and Dr. Ramadan El Shatshat, for their invaluable feedback and constructive suggestions. I am grateful to my research group Dr. Marzia Zaman, Dr. Srini Sampalli, Dr. ChungHorng Lung, Parveen Kaur, Richard Purcell, and Abdul Mutakabbir for their companionship and support. Dedication I dedicate this thesis to my parents, Aliakbar and Hooriyeh, who have always taught me tobe courageous and to step outside my comfort zone. Just as they did when they journeyed from our hometown of Janah, Iran, to Kish Island, then to Tehran, and eventually to Dubai, seeking a brighter future. Soon, they will be embarking on a new

FAQs

What problem does the thesis focus on for forest fire prediction?

It addresses the need for effective, real-time forest fire prediction and management, particularly in the Canadian context, where weather variability and region-specific risks challenge traditional methods.

Which machine learning algorithms and data source are used?

The study uses Copernicus reanalysis data and trains four algorithms: Random Forest, XGBoost, LightGBM, and CatBoost, within a feature-rich predictive framework.

How is the class imbalance handled in the dataset?

With a strong imbalance ratio (158:1 between non-fire and fire), the thesis applies re-sampling methods including under-sampling, oversampling, and hybrid techniques such as NearMiss, SMOTE, and SMOTE-ENN.

Aug 3, 2026

Upload by Lucas Martin