Credit Risk assessment is the process of investigating the probability of loss which may arise as a result of a borrower's failure to make payments of any kind of debt. Lenders such as banks stand the great risk of incurring such losses from credit or loan defaults (Sudhakar & Reddy 2016). For a loan to be refunded, there are two possible factors which may come to play: the borrower’s ability to pay and the borrower’s willingness to repay the loan. The ability of the borrower to repay the loan could largely be influenced by his or her income and other financial or economic determinants (Arya et al. 2013).On the other hand, a borrower’s willingness to repay the loan tends often to relate with certain behavioural tendencies of the borrower such as risk taking and impulsiveness.
The credit score is the commonest means used by lenders especially banks to evaluate a borrower's creditworthiness. This score is usually computed based on the borrower's historical financial records which is typically sourced from credit bureaus (Avery et al. 2000). Apart from credit bureau data, most financial institutions like credit card companies most likely have access to multiple sources of data about borrowers which could augment their lending procedures. On the other hand, behavioural tendencies like risk taking and impulsive behaviour which could influence a borrower's initiative to repay a loan appears overlooked by most financial institutions in their assessment of a borrower's credit risk. Traditionally, credit scores were computed using statistical models and techniques(Crook et al. 2007).. However, with recent developments in machine learning and artificial intelligence, there is interest by financial institutions on the improvements that novel machine learning models like Deep Learning could bring (Giudici et al. 2019).
This research aims to develop and evaluate a novel data fusion and machine learning model especially Deep Learning model for the automatic assessment of credit score. The PhD will also focus on how to combine multi-source data (with different sources, granularity, etc.) into a cohesive and accurate model, including information relating to impulsive and risk-taking behaviour about borrowers.