Prediction of loan eligibility using Machine Learning Models

Loan issuance is a very risky proposition for any bank, primarily because of the credit worthiness of the applicants. However, there are plenty of systematic and unsystematic risks those are associated with the loan issuance process. Other than the individual default cases there may be market disruptions or the factors that are beyond control for the investors like job loss, new regulations, and other extraneous factors. Banks are now becoming more prudent and extra vigilant in approving the loan applications. The final decision of approving a loan depends a lot on the approver, that means a human intervention is still required to approve a loan, but the approver can take a lot of insights from the available systems and technologies that help facilitating the entire lifecycle by automating the decision-making workflows with the help of machine learning models. Application of machine learning technology in evaluating the eligibility of a loan applicant has become more common now a days, there are plethora of technologies that can be applied to build and evaluate the eligibility criteria.

Logistic and linear regression algorithms are immensely effective in successfully carrying out such predictions. As the process suggests the system requires information specific to the person to be fed into the system as an input and the system uses that data to analyses the same and come up with the outcomes specific to the loan eligibility of the person, based on that the loan may be approved or disapproved. The key principle that the system uses here is the inductive methods for the attributes and determine the eligibility condition for the specific applicant. Under ongoing pandemic situation, the uncertainty in the job market has led to more requirements of loans and the rate of defaulters also increases. That’s why in today’s perspectives it is even more important to identify the eligible borrower using proper evaluation technique. I would like to stress on two methodologies that are used in machine learning, the first one is logistic regression, and the other one is linear regression techniques.

Machine learning is an evolving interdisciplinary field and it’s an application of artificial intelligence (AI), that augments the system with the ability to learn from existing available data set, that we refer as training data. The entire modelling process starts with the selection of appropriate data and subsequent analysis of the same. The observed data will help to get the pattern and that will help in future in enabling better decision making. The intent of this exercise is to let the system learn of its own without any human intervention so that in future the machine can take a constructive decision that will augment the holistic decision-making process. There are three types of machine learning methods, the application of what has been learned in the past to new data using labelled examples to predict future events is called supervised machine learning technique. In contrast, unsupervised machine learning algorithms are used when the information used to train is neither classified nor labelled. Semi-supervised machine learning algorithms fall somewhere in between supervised and unsupervised learning, since they use both labelled and unlabelled data for training, typically a minuscule amount of labelled data and a considerable amount of unlabelled data. Reinforcement machine learning algorithms is a learning method that interacts with its environment by producing actions and discovers errors or rewards. Trial and error search and delayed reward are the most relevant characteristics of reinforcement learning. To identify whether an applicant is eligible for loan approval depends on several independent variables. Some of which are indicated below:

The fundamental methodology of machine learning is simple, for that training data is required, the same is fed into the model, so that the model is trained on different characteristics and later real data is entered to get the output from the model.

Now a days identifying a right candidate for loan approval is much easier, the applicant needs to fill up all the necessary details and the same is captured in the system as data, the specific model is already implemented in the system. Once the model runs the data entered into the system the model comes up with the required outcome and the bank has already a threshold set up, if the parameter generated by the system has a value higher than the value of the parameter then in that case loan is approved and if the system generated parameter has value lesser than the threshold value as set up for that system in that case loan is not disbursed for that candidate.

This has also been observed over the years that a credit rating alone should not be the only deciding criteria for loan approval, there are plenty of other factors that can contribute considerably towards taking the decision whether a loan is approved or not. Borrower’s credibility depends on multiple factors, some of those are already included as part of this paper, however there are numerous others those can prove to be critical and for that a detail analysis is required. This has also been established that linear regression and logistic regression both are critical as far as the prediction of eligibility is concerned linear regression can help to identify the list of all the independent variables and the relative importance, that helps to determine which are the critical ones so that emphasis should be given to set the threshold levels for those independent variables, and anything associated with those mainly like documentations etc. should be taken care with lot of precision. The logistic regression techniques on the other hand holistically helps to take a final decision based on some binary output. From the above schematic diagram linear regression clearly indicates that the loan amount (8) and the applicant income (6) are the most important factors effecting the loan eligibility. Whereas logistic regression model is a type of generalized linear model that extends the linear regression model by linking the range of real numbers to the 0-1 range. Let’s say the deterministic value for one applicant comes out as 0.45, now if the threshold for the bank is set as 0.5, then in that case the applicant’s loan request won’t get approved as the value is lesser than the threshold value.

However, the future research should include more of such independent variables so that more detail perspectives of the analysis can be drawn to get a detail insight about the process. There is also a need to do a demographic analysis to identify the defaulters and understand the nature of the scoped transactions so that in future those can be identified and triggered at the beginning.

Thoughts, comments are welcome!