In the past few decades, forest fires have been starting to become more common and more severe. Systems for rating fire danger, such as the Canadian Forest Fire Weather Index (FWI) system, have been used in combination with machine learning algorithms to predict the severity of forest fires. In this paper, we will attempt to predict the area that will be burned by a forest fire, using a Bayesian network. Moreover, we will try to predict several components of the FWI system based on the measured weather conditions. If our attempt were successful, it could be used to improve the predictions of forest fire damage, which could have great benefits to society. However, when evaluating our results, we found some contradictory results: we obtained very poor correlations for the prediction of the size of the affected area, but our results are in line with previous work that claims to be able to predict 61% of area values when allowing an error of 2 ha.
This repository contains the code of the first assignment for the NWI-IMC012 Bayesian Networks course at Radboud University (2019-2020).
This project has the following file structure:
forestfires.csv: the dataset used for this assignment. Retrieved from the UCI Machine Learning Repository.project.R: contains the source code for the whole assignment.project.Rproj: the project settings as used by RStudio.img/: directory that contains all images generated byproject.R.old-code/: directory that contains all the non-final R files we created while working on this assignment.