Machine Learning for Ethical Credit Scoring

Steps towards building an ethical credit scoring system for low-income individuals. The following findings stem from an Omdena research project with AI startup Creedix.

The problem: Banking the unbanked

According to the World Bank, about 1.7 billion adults do not have an account at a financial institution or through a mobile money provider. In 2014 that number was 2 billion!

One of the main reasons is that “first-time-borrowers” do not possess a credit history, collateral, or any previous accounts. All of which are essential for conventional credit scoring approaches.

Applying Machine Learning for ethical credit scoring

The mission of Creedix is to build the World´s #1 Ethical Credit Scoring Solution.

One of their key value points is to provide fair and transparent scores available to everyone. All data such as financial and identity data will be fully-owned by the consumer.

The data

The team worked on three datasets that were provided by Creedix:

• Transactions made by different account numbers, the region, mode of transaction, etc
• Per capita income per area (All the data is privacy law compliant)
• Job title of the account numbers

All data was anonymous.

Engineering the features

The team decided on the following features taken from the data provided by the project partner and scrapped for additional data sources to fill in gaps. The scraped data came from numbeo to get the cost of living per area and living expenses. Data from Indeed provided salary numbers to assign an average salary to different kinds of jobs.

Correclation among chosen features

The final deliverables of the research project were the engineered features as well as machine learning pipelines both for supervised and unsupervised learning. Out of the scope of this project, is the next step where the Creedix team looks through the features and decides on the weightage for each feature to build the final machine learning models.

Detail can be found in the blog post below.