ClimiCide

• 

  Array of predicted suicides in all 50 states for Black females (2040)

Inspiration

The inspiration for ClimiCide came from a recently published EPA dataset that showed the link between temperature and suicide in the United States. I was inspired to apply machine learning principles to quantify the potential impact of climate change on mental health. As global temperatures rise, there is an increasing risk of extreme weather events, environmental degradation, and resource scarcity, all of which contribute to increased suicide rates. This is an extremely topical issue that we need to continue to bring awareness to.

How we built it

I trained the model on climate data from NOAA and suicide data from the CDC underlying death dataset. I also used predicted temperatures based on different emissions scenarios, sourced from Climate Impact Lab. The death rates were predicted suicides (as foretold by the model) divided by predicted populations, which were sourced from the St. Louis federal dataset. After building, training, testing, validating, and finally finishing the model, I deployed it onto Wix to present my findings in easily digestible interactive maps.

Challenges we ran into

Data collection was very difficult, because, as discussed on the Climate Injustice page of my website, there is not a lot of data for marginalized communities (races and genders). The CDC underlying dataset had around 20,000 records for suicide and self-harm-related deaths, yet only 19 of those records were from Native Americans or Indigenous individuals. In addition to this environmental injustice, preprocessing so many datasets to work together was very difficult.

Accomplishments that we're proud of

I am very proud of building a model with a complicated machine learning algorithm I've never used before, HistGradientBoosterRegressor. I learned a lot through this project, and it helped me translate machine learning theory to applications.

What we learned

I learned a lot about preprocessing datasets, and how to correctly format them in ways that will agree with the machine learning packages. There was a lot of manual labor involved in cleaning the datasets, because I had to make sure they were always the same data type, column name, and number of records. The skills I especially strengthened through this project were preprocessing and my familiarity with Regressor packages (I've mostly worked with classification models).

What's next for ClimiCide

I would love to add more data to ClimiCide, particularly for other races besides Black or African American and White (Hispanic and Non-Hispanic). Furthermore, being able to predict suicides in other years besides 2040 is a goal for ClimiCide.

Built With

• html5
• javascript
• machine-learning
• python
• vscode
• wix