Testing Different Kinds of Neural Networks to Analyze Patterns in Wildfires Across India

Abstract

In recent years, due to climate change and global warming, the number of wildfires has dramatically increased. This increase in wildfires has led to property damage, injury and loss of life across the world. This study aims to develop a Neural Network capable of modeling the risk of wildfires across India using weather data. This would warn firefighters before a fire starts and would lower response time, hence minimizing damage. 4 Neural Networks were tested, a simple Convolutional Neural Network, as control, a Convolutional AutoEncoder and 3 types of Recurrent Neural Networks with Convolutional layers, to capture time complexity. The Neural Networks utilized Convolutional layers to capture location complexity. 4 models achieved acceptable levels of accuracy, ranging from a Mean Squared Error Loss of under 1 to nearly 2.3.