Maximizing Prediction Accuracy in Wildfire Severity: A Comprehensive Analysis of Machine Learning Models Using Environmental Features

The escalating severity and frequency of wildfires have been widely attributed to the effects of climate change. In order to formulate efficient wildfire evacuation plans, it is imperative to examine the effect of such fires that will help in the the decision-making process involved in evacuations and ascertain the key factors that influence individuals’ choices. This paper explores the trends of wildfires in California from 2001 to 2019, focusing on the role of environmental features. It aims to predict the extent of land affected by a wildfire based on the given environmental characteristics of the fire event. The study compares and evaluates the performance of various regression and classification methods in Machine Learning, specifically considering the influence of environmental factors. Experiments were performed on real-life datasets, involving data cleaning, feature engineering, handling irregularities, and data normalization. The objective was to develop models capable of predicting the size of wildfires based on weather features in the month of ignition. While the multi-class classification models encountered challenges and exhibited suboptimal performance, the regression models demonstrated greater success in accurately predicting the number of acres burned by a fire.