Individual Creditworthiness Assessment Platform

Our goal is to develop a creditworthiness & affordability assessment model that makes decisions based on end-user financial status and not on previous credit behavior.

We will utilize the dataset provided to create a classification engine able to predict the performance of each applicant (good/bad) in a way that would allow us to embed the outcome of our existing offering. For the classification part of the infrastructure, we will use model ensembles and stacked models. Instead of a single fine-tuned model, our solution will be using various machine learning techniques (ensembling) in succession (stacking) in order to produce the most accurate predictions. The results produced by each predictor as a single model will then be fed to the next layer also consisting of other classifiers and networks. Each of these cascading layers will decrease in size in order to reach the last stage/layer which will produce the final prediction.

Experiment progress

Finclude is working on enabling those that are “credit invisible,” or those who do not meet the traditional criteria for a credit product, to get a fair shot. Getting credit is a Catch-22 for consumers with a short credit history or “thin credit file” - they cannot get credit and build their credit history, because by the default criteria they do not qualify for a credit line. Our approach uses an algorithm developed in-house that leverages machine learning to alternatively assess a consumer’s credit risk based on their transactional behaviour.

Main Results of the 1st stage of the programme: Explored and mapped people transactional patterns

  .