An explainable AI decision-support-system to automate loan underwriting

Abstract

Widespread adoption of automated decision making by artificial intelligence (AI) is witnessed due to specular advances in computation power and improvements in optimization algorithms especially in machine learning (ML). Complex ML models provide good prediction accuracy; however, the opacity of ML models does not provide sufficient assurance for their adoption in the automation of lending decisions. This paper presents an explainable AI decision-support-system to automate the loan underwriting process by belief-rule-base (BRB). This system can accommodate human knowledge and can also learn from historical data by supervised learning. The hierarchical structure of BRB can accommodates factual and heuristic rules. The system can explain the chain of events leading to a decision for a loan application by the importance of an activated rule and the contribution of antecedent attributes in the rule. A business case study on automation of mortgage underwriting is demonstrated to show that the BRB system can provide a good trade-off between accuracy and explainability. The textual explanation produced by the activation of rules could be used as a reason for denial of a loan. The decision-making process for an application can be comprehended by the significance of rules in providing the decision and contribution of its antecedent attributes. Manual Underwriting Challenges: Manual underwriting task is a very much paper-based process. It is an inconvenient process of circulation of loan application files within different departments of a lending institution. Full attention to details is requisite to give sound judgment on an application. Human underwriter evaluates scenarios by analysing a large amount of dynamic information in a loan application. This could be a source of inconsistency, inaccuracy, and biases (Peterson, 2017). Manual underwriting is often successful in processing non-standard loans. Many lenders like high street banks and building societies follow strict rules and do not offer personalized underwriting. However, there are some lenders who exercise common sense lending approaches for assessing both standard and nonstandard cases such as non-standard properties, non-standard income/employment, and less than perfect credit scores. The underwriting process of non-standard cases is very detailed and individualistic. Common senses lending approaches serve underserved people Advantages and Concerns of Automated Underwriting Systems: In 1995 Fannie Mae, a US largest mortgage lending company introduced first Desktop Underwriter that applied both heuristics and statistics to process mortgage loan in less time, cost, and paperwork (Cocheo, 1995). In the same year, another US lending company, Freddie Mac introduced an automated underwriting system called loan prospector (Cocheo, 1995). Most lenders use an automated system which contains coded underwriter guidelines which provide the decision of acceptance or rejection when certain default rules in the rule base are triggered. Early statistical methods were limited t...

Artificial Intelligence and Bank Soundness: Between the Devil and the Deep Blue Sea - Part 2

“…As such employ verification and checks at several levels to weed out incorrect or weak decisions. As such, Loan officers should be able to provide a logical explanation on what grounds a loan has been accepted or rejected to their superiors, compliance officers, auditors, regulators and customers [5,7,10,12,64,65]. The working logic of AI decision has to be traceable backwards.…”

“…Transparency is also important to fully trust the system through validating the decision made by AI, by not only detecting anomalies in the decision process such as biasness, mistakes, manipulations of data, deficiencies, compliance to rules i.e. GDPR, cybersecurity crimes linked to work processes such as dataset poisoning, internal network manipulation, and side-channel attacks [69] but also to detect clearly and precisely at which step the anomalies occurred and what information AI fed itself [10,12,64,[66][67][68][70][71][72].…”

“…Yet, to train AI systems to replicate human decisionmaking skills is a challenge. As it is difficult to transform various algorithmic concepts into training data to solve every problem for a range of lending products [10,77].…”

“…The chapter contributes to literature in several ways. Earlier research has either focused on AI application on the entire financial sector covering banks, fintech companies, mortgage lenders, security companies amongst others [5][6][7] or have evaluated AI applications in the form of Machine Learning (ML), Neural Networks (NN) Artificial Neural Networks (ANN) in specific areas such as credit evaluation, portfolio management, financial prediction and planning [2,[8][9][10] or by examining user friendly experiences of end users [5,6,11]. Therefore, these studies are not sufficient to understand the challenges proposed by AI from solely a bank's perspective.…”

Are incumbent banks bygones in the face of digital transformation?

“…There is a plethora of behavioural data available; even the friend's list on Facebook can be used as an assessment criterion (Jarrahi, 2018). These models should enhance explainability and interpretability besides accuracy (Sachan et al, 2020) in order to outperform the existing mechanisms. There are obvious benefits both for the banks that can make better and more granular credit decisions, and for the customers that receive services in a transparent and timely manner.…”