V3I5P42

Development of a Loan Decision System Using Multilayer Perceptron and K-Nearest Neighbor Classifier

Michael Favour Edafeajiroke^1*, Daniel Umejuru²

Abstract

The manual, subjective nature of traditional loan approval poses significant risks to banking stability, driving the urgent need for the adoption of automated, data-driven systems. While machine learning offers transformative potential, a critical gap persists between the high accuracy of complex “black-box” models and the practical need for interpretable, robust decision-support tools. This study addresses this gap by empirically evaluating and comparing two machine learning classifiers, the K-Nearest Neighbor (KNN) algorithm and a Multilayer Perceptron (MLP) neural network for loan default prediction, utilizing a real-world dataset and Chi-square feature selection. The results demonstrate a clear trade-off: the MLP achieved superior predictive accuracy (94.72%) and better detection of default cases, whereas the KNN, while faster and more interpretable, attained lower accuracy (84.80%) and poor sensitivity. The findings substantiate that effective credit risk modeling must balance analytical power with operational practicality, highlighting the necessity for future hybrid systems that integrate interpretability with high performance for reliable deployment in financial institutions.

Keywords:

Credit Risk Modeling, Machine Learning, Loan Default Prediction, Predictability, Multilayer Perceptron, K-Nearest Neighbor.