KNN Based Big Data Analytics Framework for Real-Time Business Decision Support

Abstract

Credit risk assessment is a critical aspect of financial decision-making, requiring accurate models to predict the likelihood of loan defaults. Traditional statistical methods often struggle with complex, high-dimensional datasets, making machine learning (ML) techniques a preferred alternative. This study evaluates multiple ML algorithms, including Decision Tree, Support Vector Machine, Adaboost, Random Forest, XGB Classifier, and a proposed K-Neighbors Classifier, to determine their effectiveness in predicting credit risk. The results indicate that the Proposed K-Neighbors Classifier achieved the highest test accuracy (85.25%), but with a higher Mean Squared Error, suggesting potential overfitting. Among other models, Random Forest and XGB Classifier demonstrated strong generalization with balanced accuracy and lower error rates, making them suitable for practical credit risk assessment. The comparative analysis with existing systems highlights the advantages of ML models over traditional approaches in handling large credit datasets. The study emphasizes the need for optimized algorithms that balance accuracy and robustness while minimizing overfitting. Future research should focus on hybrid models, deep learning techniques, and alternative data sources, such as behavioral and transaction-based data, to enhance predictive performance. By leveraging advanced machine learning techniques, financial institutions can improve credit risk evaluation, leading to more informed and efficient lending decisions.