Building Reliable Loan Approval Systems: Leveraging Feature Engineering and Machine Learning

Abstract

Automating loan approval system is essential in today's banking system.  Even with the shift to online platforms, the traditional method is still cumbersome and needs a lot of customer-related data. This study proposes a robust solution to overcome these challenges. Despite previous studies, new financial indicators in feature engineering stage are introduces to extract more important client information, thereby improving prediction robustness and accuracy. To implement our integrated approach, an online dataset from a finance company, is utilized. The dataset is preprocessed by various data preparation techniques, including cleaning, transformation, and feature engineering. Subsequently, the preprocessed data undergoes a range of powerful machine learning techniques such as K-Nearest Neighbor, Decision Tree, Gaussian Naive Bayes, and Logistic Regression. Additionally, three robust ensemble methods including Random Forest, AdaBoost Classifier, and Gradient Boosting Classifier are employed for further improveness in performance.  The presented approach succeeded to acheive the highest accuracy by AdaBoost Classifier at 88%. A comparison with the original preprocessed model using ROC curve and feature importance analysis demonstrates the superior performance of our approach, with a larger area under the ROC curve and reduced false positive rate. Furthermore, the comparison findings show a stronger reliance of our model on financial features rather than personal customer features, highlighting its robust classification performance. These results indicate the potential strength of our model to replace the current loan approval system in real-world applications.