What it does
We generated a website where a user can calculate your "carbon score" - similar to a credit score - using a variety of parameters such as the area you live in, its climate, estimated household income if you have a pool/bath, etc. Using this carbon score we predict how one can improve their sustainability indicators.
How to Run
Please go to carbon/templates and run the index.html file to launch the website locally. The user must enter the fields requested in the website and click "Run" to get redirected to a webpage with their carbon score and other relevant statistics. The user can then also find how they fare against other users and fairness data to see how sustainability scores as skewed to favor certain demographics due to inherent biases.
Inspiration
Climate change has become an inevitable catastrophe that we as a human race must take accountability for. Despite large corporations and countries contributing heavily to greenhouse gasses and large carbon footprints that create ecological disorder, individual contributions are significant, especially in places disproportionately affected by such activities. Since data regarding people's homes, incomes, and lifestyle choices are available through surveys on the US Census Bureau, we decided to leverage this information to predict how a user may fare against others within their income strata or against other groups and gives suggestions on how to improve their footprint.
How we built it
We trained 18,000 data points from the US Census Bureau describing household habits and features from all 50 states. We ran a SARIMAX model (which creates a score based on trends from moving averages) to give each of these households a representative score compared to other data points. With this scaled score, we compared it to other factors, which, after thorough research, we decided contributed heavily to a person's sustainability index. We then compared this score to these generalizations about a person's sustainability choices. We used confusion matrices to see if our scaled carbon score accurately predicted people's sustainability and lifestyle choices.
Challenges we ran into
Training the model was difficult as we had to heavily preprocess our data and find features that contributed significantly to the model's error and overall accuracy. We had to look through various models from several sources such as toolkits to find algorithms that reduced our RMSE error and provided not only accurate predictions but also fair predictions across demographics (we checked for low errors for both high and low-income communities and regions across the United States).
Accomplishments that we're proud of
We were able to achieve a high overall accuracy across the strata, we made meaning that the model is a good first step in helping users contribute towards their individual sustainability initiatives. However, the error for lower-income families was much higher at (60%) compared to (90%) for high-income families. This may have been due to high-income families being more likely to respond in such surveys, hence we need to ensure data is unbiased and we have more than merely 18,000 data points to run our ML algorithm.
What's Next
We believe we do not have enough data points which is why we may observe a higher error for certain subgroups. We also want to make the Website more user-friendly to help users interact and feel more comfortable. interacting with the platform.
Built With
- full-stack
- javascript
- node.js
- python
Log in or sign up for Devpost to join the conversation.