Using Machine Learning and Aggregated Remote Sensing Data for Wildfire Occurrence Prediction and Feature Selection: A Case Study in California

Due to global warming, wildfires are becoming increasingly frequent and destructive, threatening environmental, economic, and human well-being on a global scale. Recent advancements in remote sensing and advanced data analytics have spurred the development of fire occurrence prediction models (FOPMs) to tackle this challenge. Although a plethora of features have been employed in the development of FOPMs in prior studies, identification of the most relevant features and optimal feature subset remains a critical knowledge gap. Utilizing California as a case study, this study fills this knowledge gap by conducting a comprehensive investigation on 96 relevant features gathered from seven heterogeneous databases. Ten machine learning algorithms were tested and employed with four feature importance methods to derive an importance score for all the features. Eleven features were identified as the optimal feature subset, and XGBoost achieved the best prediction performance with F-score of 97.35%.