Predicting the Danger of Particulate Matter Pollution from Wildfires Using Classification Models

Abstract

               Due to a mix of climate change and California’s mega-drought, California’s wildfire seasons have overall gotten progressively longer, more destructive, and more expensive. In 2020 alone, around 9,900 wildfires burned about 4.3 million acres, costing the state over $12 billion. (Kerlin, 2022) Larger and more numerous wildfires pollute billions of harmful particles into the atmosphere, including PM2.5. This study aims to use features of a wildfire and other factors to predict whether a wildfire pollutes enough PM2.5 particles to be detrimental to human health. The 8 features used in the model are the acres burned, the length in days of the fire, available green space within a 15-mile radius of the fire, the highest population density within a 15-mile radius of the fire, electricity usage, median income, temperature, and precipitation. A Gradient Boosting Classifier (GBC) was applied to the dataset to predict whether a wildfire’s emissions necessitated an evacuation. The GBC results achieved a high accuracy of 0.931 as well as a great Area Under the Curve (AUC) of 0.911. By far the most important feature in the GBC is Length, with a feature importance score of 0.109 +/- 0.009.