Enhancing credit score classification: a time-effective deep learning approach

Abstract

This study aims to evaluate the performance of multilayer perceptron (MLP) deep learning models in credit scoring, focusing on both accuracy and execution time. By analyzing datasets with up to 100,000 instances, we assess various metrics, including Accuracy, F1-score, Recall, Precision and processing speed. Special attention is given to the LBS-MLP model, which achieves significantly faster execution times (2–4 times faster) without sacrificing accuracy, offering practical benefits for credit risk management. These insights aim to support financial institutions in making timely, efficient and informed lending decisions. This study employs a comparative analysis of various MLP deep learning models for credit scoring, using datasets of up to 100,000 instances. We assess model performance based on key metrics, including Accuracy, F1-score, Recall, Precision and execution time. The models are trained and tested using cross-validation techniques to ensure robust evaluation. Special focus is given to the LBS-MLP model, designed to improve execution time efficiency without compromising accuracy. The research framework emphasizes practical applicability, offering insights into optimizing credit risk assessment in financial institutions. The study reveals that while most MLP models achieved similar accuracy levels in credit scoring, the LBS-MLP model stands out for its efficiency. LBS-MLP reduced execution time by 2–4 times compared to other models, without compromising accuracy. For instance, with a dataset of 100,000 instances, LBS-MLP completed tasks in 62.9682 s, significantly faster than its counterparts. In contrast, models like Baseline MLP and HA-MLP, though accurate, did not match the time efficiency of LBS-MLP. The findings highlight the potential of LBS-MLP for improving both accuracy and speed in credit risk assessment. This study introduces the LBS-MLP model, which enhances MLP deep learning for credit scoring by significantly improving execution time without sacrificing accuracy. It provides a novel approach to balancing efficiency and performance in credit risk assessment, offering practical value for financial institutions. The research fills a gap by evaluating MLP models on both performance metrics and execution time, which is often overlooked. By demonstrating the LBS-MLP model's superior efficiency, this study contributes valuable insights for optimizing credit scoring processes and supports informed decision-making in financial risk management.