Master of Science in Business Analytics
Master of Science in Business Analytics
Home
About
Program
Courses
Faculty
Application
Assistantship
Events
Projects
Resources
FAQ
Projects
2021-2022
California Wildfire Prediction Using Ensemble Machine Learning Methods
Student Team
Thomas Logue, Yixuan Song, Nabila Zia, David Wenyu, Christian Gonzales, Huy Tran
Faculty Advisors
Dr. Mohammad Salehan, Dr. Shuo Zeng
Purpose
This thesis aims to explore machine learning models and their effectiveness in decreasing the overall impact wildfires have on our environment in the western United States. By utilizing publicly available weather, fire, and remotely sensed data we can predict certain characteristics of a wildfire that can reduce the impact and costs that wildfires have on our communities.
Study Design/Methodology/Approach
Our research uses Machine learning, deep learning, and computer vision models to aid researchers in predicting climate related events by leveraging vast amounts of historical weather and remotely sensed data. This thesis demonstrates the effectiveness of using these models to predict the probability of a wildfire occurring within a 100km-by-100km area and a 20m-by-20m area and illustrates how the models were developed and applied using this data.
Findings
Our results were validated with confusion matrices, Receiver Operating Characteristics curves (ROC), and loss functions. In addition, we projected our predicted fire locations onto a map and found our models could predict a fire occurring within 1.5 miles of the actual location.
Originality/Value
Wildfires and climate change related events continue to have a significant impact and cost on people and local communities in the state of California. These models and methods could give firefighters and mitigation planners more precise future fire locations reducing the overall impact and cost of wildfires.
Practical Implications
These results may be used to increase the effectiveness of current models used to forecast and predict wildfires by providing additional information to interested parties or stakeholders. Additionally, these methods could be used to forecast or simulate other climate related events that occur frequently.
Keywords
wildfire, prediction, sentinel-2, remote sensing, weather, logistic regression, deep learning, random forest