Our customer, a mid-size bank in the United States, was finding it difficult to process loan applications received from various regions. In addition, their approval decisions were often governed by subjective criteria such as credit score and personal history. To streamline this tedious task, they sought an automated intelligent loan application solution that would consider multiple factors before recommending whether or not to grant the loans requested.
Business Challenges
Mundane activities were injected into loan application eligibility checks, and as a result, decision-making processes and approval workflows were hindered. Monthly reports required painstaking manual data preparation that consumed too much time, which led to further challenges in maintaining high levels of data quality due to a lack of an efficient cleansing framework. Compounding this problem was the absence of any prediction setup for determining qualification status. The client wanted to automate the loan application eligibility review process and reduce the time spent on monthly reporting. The challenges can be summarized as follows:
- Manual loan application eligibility process delayed approvals and decision-making.
- Monthly reports required time-consuming, manual data preparation.
- An inefficient cleansing framework brought down data quality levels.
- Absence of prediction setup system to determine qualification status.
Benefits
- ML model replaced the manual loan application solution
- Loan eligibility decisions in minutes instead of days or weeks
- Proper data cleansing and preparation for monthly reporting using Power BI
- Ability to predict eligibilities and loan amounts
- Automated approval workflow improved decision-making
PreludeSys Solution
Through comprehensive exploratory data analysis, we designed eligibility prediction models and balanced them using the Synthetic Minority Oversampling Technique (SMOTE). We then performed an analysis to create ML models using technologies such as Gradient Boosting, Logistic Regression, Random Forest, and XGB Classifier. Hyperparameter tuning further improved ML model performance which was evaluated through Accuracy and Area Under ROC Curve (AUC) metrics. Power BI visualizations depicting loans status (approved/rejected) by reason provided valuable insight into further understanding their results.