An integrated resampling and machine learning framework for predictive analytics of large wildfires

Accurately predicting the final burned area of wildfires at the moment of ignition is a challenging problem with important implications for disaster management and resource allocation. In this study, we apply recently developed resampling techniques, combined with machine learning algorithms, to improve predictive accuracy for rare, high-impact wildfire events. While the study focuses on wildfires, the methodology addresses a broader challenge in decision analytics: generating reliable predictions from imbalanced datasets to support informed managerial decisions. Our results demonstrate that these techniques can enhance the robustness of predictive models for extreme events, offering insights that may be relevant to other domains where accurate forecasts of rare outcomes are critical. • Propose an integrated resampling and machine learning framework for predictive analytics in large wildfires. • Apply resampling techniques to address imbalanced wildfire ignition data. • Enhance predictive performance of rare and high-impact wildfires with imbalanced datasets. • Provide decision support for wildfire management and resource allocation planning. • Demonstrate the broader value of predictive analytics for extreme event forecasting.