Predicting the Impact of Wildfire Using Machine Learning Techniques to Assist Effective Deployment of Resources

Wildfire incidents occur every year due to human-caused factors, droughts, vegetation, and climate change. It is complex to predict severe wildfires and even more difficult to extinct them. Structural damage may occur during wildfires incidents. Therefore, understanding the intensity of these damages could help mitigate the risk for the region and local communities. This study builds a predictive model using various Machine Learning techniques as a real-time decision-support tool. The model predicts the severity of damage to the structures affected during the wildfires in the state of California, United States. The study used The California Department of Forestry and Fire Protection (CAL FIRE)'s historical data from 2013 until 2019. The study trained several supervised machine learning algorithms for classification, such as Logistic Regression (LR), Decision Trees (DT), Random Forest (RF), Support Vector Machine (SVM), Stochastic Gradient Descent (SGD), Naïve Bayes (NB), and XGBoost (XGB), to predict the structural damage. Predictions were classified in the following categories of expected Fire damage: Inaccessible, Destroyed (>50
^%
), Major (26-50
^%
), Minor (10-25%), Affected (1-9%), and No Damage. The best performing algorithm was the Decision Tree model. This study will allow fire and forest departments, other departments, corporations, and the public to forecast the expected fire damage when the wildfire is first reported. It will also enable the departments, corporations, and public to prioritize actions and deploy their resources effectively to reduce the impact of wildfires in the region.