The inspiration for our project came from our connection to economics. Some of us have taken classes and the idea of GDP comes up with great importance. So we thought it would be an interesting project to predict GDP growth with several variables.
In 2008, the members of this team were in 1st grade learning addition and subtraction while the Great Recession was sacrificing some of their parent's jobs. Now, in 2021 as COVID-19 continues to pull the USA into a recession, we as college freshman armed with the power of machine learning hope to help reduce the destructive power of economic recessions.
What it does
Typically the business cycle is analyzed in retrospect as looking into the past allows an analyst to see the larger picture and label economic periods. Our model aims to analyze these changes in the business cycle, allowing preventative measures to be enacted by government agencies to prevent economic downturn. With a strong, predictive, data-driven model, the job of recognizing recession becomes easier and facilitates timely responses by these agencies.
How we built it
Utilizing the databasing capabilities of Django, a framework for data storage and retrieval was designed for user ease of use. At the click of a checkbox menu, a user has the power to select any and all data they wish to compare and contrast with no significant time lag for large datasets. Utilizing scikit-learn and linear regression machine learning, we trained a model on data collected by the Federal Reserve Bank of St. Louis. GitHub was integral to the workflow of this project, allowing developers to split workflow into independent tasks such as machine learning, frontend, backend, and data visualization.
Challenges we ran into
Due to the complexity of storing python archive files (.pkl, "pickle") in a Django database, the database is underutilized. This was circumnavigated through the use of static files. Finding correlations in novel datasets is as challenging as it might seem. Working with a smaller dataset, carefully interpreting and training the model was imperative to create a viable, real-world product.
Accomplishments that we're proud of
A working frontend and backend - not only are they both functional, they are tightly integrated and work well together. A few novel correlations were identified in our model, warranting further investigation. Advanced graphical features, such as interactability, was implemented to visualize an otherwise abstract machine learning model.
What we learned
Throughout the course of this hackathon, we all learned more about economics and predicting growth. For example, our machine learning expert gained an appreciation for business, while our economics expert gained an appreciation for how websites he interacts with every day are created. In a high-stress, time-constrained environment, we all learned how to better collaborate and communicate our ideas.
What's next for NET GAIN
We hope that through future conversations with economic and financial experts (perhaps even at this event!), we hope to learn more about the problem space, and hopefully refine, redesign, and realign our model and user interface. At the end of this project's lifecycle, which we anticipate supporting as long as possible, we hope to have a tool that professionals can use from academia to Wall Street.