Back To Projects

The Motions of Forest Fires

Giuliano T.

Can I make an AI model that can predict how forest fires will move? I am trying to take data about the environment and a current fire and predict where the fire will go / if it will grow or shrink.


Can I make an AI model that can predict how forest fires will move? I am trying to take data about the environment and a current fire and predict where the fire will go / if it will grow or shrink. Forest fires are burning more and more each year, as evidenced by the record breaking fires in Australia in 2020. The top ten years with the largest annual acreage burnt have all happened since 2004, and forest fires are now causing 7.4 million more acres of tree cover loss per year since 2001. This project would hopefully let firefighters predict where the fire is going to prevent it. The data I used was a graph of where the fire was and other features such as the surrounding wind speed, terrain, plant content, etc, and then the label comes from another graph showing where the fire moved to. I tried several different models, namely a Multi Layer Perceptron (or Neural Net), Logistic Regression and Linear Regression before settling on a CNN that predicts if each square kilometer will have a fire or not. I then was unable to get this to work so I settled on making a model that predicts how dangerous the fire will be. I was eventually able to get it to predict this with 60% accuracy. The major conclusion I had with this problem is that it is very tough but it seems like there is a way to predict where a forest fire will go. I would like to keep researching to see if I can find out how.

Explore More!

Giuliano T.
Sophia Barton
Computer Science MS from Stanford

Related Projects

One Class Classification for Overdose Death Detection

In the past year alone there were an estimated 107,622 drug overdose deaths in just the United States. With such an incredible amount of deaths from just this, I thought it would be beneficial to create a model to predict who is at risk of drug overdose deaths.
Ram N.
Mentored by Eric Bradford
workspace_premium
Differences in predicted rates of vaginal births after cesarean across racial groups in a ‘race-neutral’ model

A large body of work in machine learning has highlighted that supposedly de-biased systems often re-code sensitive variables like race in terms of proxy variables. In order to determine if this was the case in this calculator, we replicated their formula, then found base-rate statistics of all the input variables for three different racial groups: Black, White, and Asian.
Anjali S.
Mentored by Katie O'Nell
workspace_premium
A Machine Learning Approach to Understanding the Determining Factors of the Gender Wage Gap

By studying the affect of different attributes on the gender wage gap, we can better understand both the scale of this issue and its possible solutions. So, we explore the question, how does a worker’s marital status, along with other variables, impact the gap in hourly wage between male and female workers? We seek to create a model able to predict the gender wage gap given a set of variables—age, years of education, race, state, and marital status.
Sophia G.
Mentored by Ana Sofia Muñoz Valadez