An Experimental Study on Credit Score Prediction Using K-Nearest Neighbors

Abstract

Credit score classification is a vital component of risk management in the financial sector. It conventionally relies on obsolete models that fail to capture dynamic patterns. This study utilizes K-Nearest Neighbor (KNN) on a dataset comprised of demographic and financial features. The methodology involved preprocessing steps, followed by model training and performance calculations. The KNN model produced an accuracy of 88%, which exceeded the desired threshold of 70%. However, performance varied across the classes, with Low credit scores demonstrating lower precision and F1-scores. This shows the model had difficulty in predicting minority classes. The correlation matrix showed feeble linear relationships amongst the features, which significantly limit predictive power. The results of this study demonstrate strong potential for credit scoring but should be explored with substitute models and feature engineering to provide precision and impartiality in credit score classification.