Enhancement of Digital Credit Scoring in P2P Lending using a Robust Hybrid Voting–Stacking Ensemble Framework

[Purpose] This study evaluates a hybrid ensemble framework designed to enhance credit-risk prediction accuracy in peer-to-peer (P2P) digital lending platforms and to overcome the limitations of existing models. [Design/methodology/approach] The proposed Hybrid Optimized Ensemble Learning (HOEL) framework integrates seven base classifiers through a three-level optimization process that involves K-fold cross-validation, hyperparameter tuning, an intermediate weighted-voting ensemble layer, and a final stacking meta-learner. [Findings] Empirical results demonstrate that the HOEL framework outperforms individual classifiers, achieving an ROC-AUC of 97.62% and an accuracy of 96.15%. The ensemble’s layered design improves predictive stability and interpretability, confirming its robustness across small and high-dimensional datasets. [Research limitations] Although computationally intensive, the framework’s performance can be further optimized using cloud-based or parallel processing. Future studies could incorporate additional behavioral features to improve model generalization. [Practical implications] The framework can be integrated into a trusted P2P digital lending platform to improve loan-approval decisions for customers with limited credit history, thereby reducing lending risks. [Originality/value] HOEL’s novelty resides in its three-level optimization architecture, which embeds an intermediate weighted-voting layer that stabilizes ensemble predictions before they are passed to a stacking meta-learner. This combination has not previously been applied to imbalanced P2P credit risk scoring and delivers measurable gains in both ROC–AUC and accuracy over conventional ensemble approaches. The voting layer reduces hyperparameter sensitivity and strengthens stacking stability, making the framework particularly effective for small and high-dimensional financial datasets. This study thereby advances the Decision Sciences literature on quantitative risk modeling by providing a replicable, data-driven credit-assessment framework that supports more informed financial decision-making under uncertainty.