The aim of this project is to support small businesses in the United States that form the backbone of the economy. These small businesses are hit the worst and they need support to get back on their feet again. One of the major factors contributing to their growth is the financial aid from banks in the form of loans but due to the COVID-19 pandemic situation, banks have become reluctant to grant loans since they are whether the small businesses will make any profits and repay the loan amount. The dataset used is Small Business Administration which contains Non-Covid related factors and external demographics data is used to calculate the probabilities of a business getting successful during these harsh situations.
What it does
There were no datasets readily available which modeled the probabilities considering the COVID-19 situation. To predict the probabilities during these times, a few more parameters are added to the dataset. These parameters are per capita deaths, unemployment claims, Twitter hashtags, Google waiting time, and Instagram posts. These parameters are used to build a new algorithm that will adjust the probability of default. For example, as the number of COVID-19 cases in a particular area increases, the algorithm will add some weight to the prior probability and calculate the new predicted value.
How I built it
Challenges I ran into
The main aim of this project is to not use the usual fraud detection values which have not considered COVID-19 repercussions on their existing data. We had to fetch the right data and perform analysis such that banks take into account the current pandemic situation and then deciding on providing loans. It was also important to understand the mathematical reasoning behind building the algorithm to predict the probability of default. Apart from this, we faced another technical issue where the laptop of one of the team members stopped working. This hardware problem led to some setbacks but we hustled through it and completed our planned work within time!
Accomplishments that I'm proud of
There was a steep learning curve that helped to apply the existing knowledge to a practical problem. This problem had to be addressed as soon as possible and we got the opportunity to work on it and support the local businesses using technology. No one from our team has a financial background but we tried our best to interpret this financial data and provide findings in the form of technology.
What I learned
This project was a great learning curve that enabled us to understand the statistical reasoning behind the usage of machine learning algorithms. Datathon also gave everyone a chance to collaborate, exchange ideas, and make attempts at newer challenges.
What's next for DEFAULT PREDICTION 2.0: #SUPPORTLOCALS
In the future, this model can be developed even further when better datasets become available for analysis. Also, the current model works on County-level data in each state. This granularity-level can also be changed to obtain an in-depth depth of each local area.