by Charity Delmus, En-Cheng Chang, Jack Nicholls, Laoibhse Nifhaolain, William Blanzeisky, Zahra Safavifar
The 2021 ML Labs Summer School challenged the students of ML Labs to create a concept using machine learning for social good. The winning group, consisting of six ML Labs students, proposed a micro-lending web application focusing on giving access to underserved populations across Sub-Saharan Africa to green energy financing. The micro-lending would be footed through peer-to-peer lending.
This idea was birthed through hours of brainstorming. The ML Labs Summer School was fortunate to receive brainstorming and creativity training delivered by the flowgroup. Accompanying this training was an ethics canvas provided by UCD which allowed us to apply our understanding of the impact of our potential project across multiple areas including the wider social and economic impact of our project, the unique value proposition, the beneficiaries, and funding. This canvas was an important step for us in iteratively visiting our idea against the PowerPoint presentation we were creating. Included in our work was a customer roadmap which helped visualise and neatly capture the flow of information we would require, what the customer needs to do, what the investors would receive, and how the repayment structure would work.
The team had initially landed on a micro-lending idea to give underserved people in the world access to credit. As this is a broad concept, we narrowed down the target audience to Sub-Saharan Africa, as we were familiar with the current business transaction methods which are performed via mobile phone by transferring phone credit. The reason we honed in on green energy loans was to tie the social good to a climate impact on top of developing areas of poverty by granting them access to a power grid. Currently, there is 590million people in Africa who do not have access to electricity. To further complement this project, the Africa Development Bank Group (ADBG) are directly investing in businesses with green power strategies, which our idea would fall under. Direct engagement with telecom companies is absolutely necessary in extracting information on customers including their mobile phone credit transaction history, this also aids in the targeting of individuals through data analytics to ensure we are targeting the right customer.
The machine learning proposed in this project would score card the potential customers, and output a probability of default for the investors to fully understand their risk. There is very recent research in the area of reinforcement learning being applied to micro-lending for the purpose of generating a probability of default. Score carding is a traditional banking method used to determine the credit risk on a potential customer. Popular methods include random forests, and logistic regression. This credit risk can determine the interest rates charged on the principal loaned. The interest rates on repayments offered to the customers of the website would be lower due to the loss in return for investors being footed by the ADBG investment grants. The process would ideally be very transparent for both parties in the business transaction, as certain risks are usually privy to a financial institution.
We would love to see this idea being followed through as it is very unique and not currently offered anywhere in the world.