Introduction
The company’s objective is to study the bank’s customers’ data to examine the factors that can result in loan default and develop a machine learning (ML) algorithm that can predict the likelihood of loan default by an applicant. The company thinks it is crucial for its future success because the algorithm can assess applicants’ loan default risk. They intend to know how to conduct the loan verification process to minimize defaulting.
The goal of the analysis was to determine and rank the factors that lead to loan defaulting based on the data provided. In the process, I had to answer questions such as what factors are associated with loan defaulting by applicants? Is it possible to predict, and how accurate is it? How many costly errors can the model produce, and are there any actions or policies that the bank can implement to reduce loan default? These questions matter because the accuracy of the algorithm shall determine whether its application is useful to the bank or whether it will cause problems such as costly errors.
Exploratory Data Analysis
The findings show that approximately all factors affected the outcome of the loan default and that interest rate played a major role in predicting the outcome of loan default as compared to other factors—the determination of the factors that affected the loan defaulting outcome involved querying. Considering loan purpose as a factor, the default rate was higher when applicants borrowed the funds for credit card and medical purposes by approximately 64% and 54%, respectively. In a similar scenario, interest rates were factored in, and in some cases, the default rate was 100%, while in other cases, it was 0%. The analysis’s findings are important for the business so that the business can determine whether the machine learning model can be used to conduct its loan processing.
The machine learning algorithm used was linear regression, where the R-squared value was determined to be 0.526 while the root mean squared value was found to be 0.334. The R-squared value measures the relationship between the actual and predicted values. It ranges from 0 to 1, where 0 signifies no relationship while 1 signifies a perfect relationship. On the other hand, the root mean squared value measures the error of the model, and in this case, the RMSE is closer to zero.
ML Model Building Exercise
The initial step involved determining the differences in loan default rates by various variables to know the variables that would be included in the model. The process involved calculating the number of customers who applied for the loan, the number of customers who defaulted, and the percentage rate for each variable. The model’s performance was measured using the AUC (Area Under ROC Curve). The AUC was 0.936873, indicating that the model was approximately 93.6873% correct. The AUC-ROC curve performance metric is used in classification problems to measure probability and degree of separability.
While the ROC measures probability, the AUC measures the degree of separability. When AUC is 0, it means the model is poor and it is worse at measuring separability. When it is 0.5, the model does not provide any class separability. The higher the AUC, the better the model is at predicting 0s and 1s classes (Narkede, 2018). For example, the model shows that the loan default is expected to be either yes or no. Since the AUC result is greater than 0.5, this indicates that it is better at distinguishing whether the loan has defaulted or not.
The variables used in the prediction included interest rate, loan term, installment, loan amount, loan purpose, home ownership, missed payment years, and debt-to-income ratio. The importance graph showed that the interest rate was the leading factor in loan defaulting. This was followed by factors such as loan term in years, installment, and loan amount, which appeared to have a higher impact on loan default by applicants.
Recommendations
The algorithm was approximately 93% correct, showing that the company can use it to predict and minimize loan default rates. To minimize the chances of defaulting, the company can implement policies that govern its decisions for interest rates, loan terms, installments, and loan amounts. For example, they can attempt to reduce the interest rates, loan duration from five to three-year terms, installments, and the loan amount that they offer to the applicants. Since the AUC is not exactly 1, it shows that the model is not 100% accurate, and therefore, costly errors can result from it.
My recommendation is that it is not possible to predict whether or not a loan applicant can default on loan repayment. One reason is that everyone has their own decisions on whether or not to repay loans, and people change decisions rapidly, so decisions cannot be predicted easily. Everyone has a unique way of life that is different from that of another person. Secondly, no one knows what the future holds but our Sole Creator, and therefore, we should focus only on the present.
Conclusion
In conclusion, the analysis results show that the variables used within the dataset provided could affect the defaulting rates of applicants. By making changes to the bank policies regarding the variables of the loan lending process, they can control how loans default. Since the model is closer to 100% accurate, it can be used to analyze loan applications and reduce defaulting on loans. Both technical and fundamental factors need to be looked into for a more accurate decision to be made.
Reference
Narkede, S. (2018). Understanding AUC-ROC Curve. Towards Data Science. Web.