Achieving Performance Trade-offs in Imbalanced Credit Default Prediction

DOI

10.2991/978-94-6239-801-6_139How to use a DOI?

Keywords

Credit Default Prediction; Imbalanced Classification; Logistic Regression; Random Forest; Ensemble Learning

Abstract

This study explores the challenge of achieving a balance in performance when predicting credit defaults on imbalanced datasets. In this paper, a probability-weighted ensemble model is proposed to integrate the advantages of the two basic models, i.e., logistic regression and random forests. It studied in depth the performance of several models on imbalanced datasets, analysing data pre-processing, features, and multi-dimensional metrics. Based on this, some optimisation approaches such as dynamic gating hybridisation and residual-enhanced hybridisation were proposed and verified the balancing performance by combining them with the balancing strategies such as SMOTE and random oversampling. Finally, the robustness of the results was validated by five-fold hierarchical cross-validation. The results show that, as far as recall is concerned, the best model is logistic regression, while as far as precision is concerned, the best model is random forests. The hybrid model did not offer a substantial improvement in any one metric but showed stable and good performance in the F1 score, a composite metric, indicating a superior capacity to manage the trade-off between precision and recall. Moreover, for the different optimisation approaches, complicated models and optimisation strategies did not lead to significantly improved balancing performance. In summary, the hybrid model’s strength has never been in optimising a single metric, but in the synergistic optimisation of many metrics; blindly increasing model complexity cannot significantly boost the balancing capacity for unbalanced classification.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Guanzhi Li
PY  - 2026
DA  - 2026/10/23
TI  - Achieving Performance Trade-offs in Imbalanced Credit Default Prediction
BT  - Proceedings of the 2026 2nd International Conference on Financial Innovation and Marketing Management (FIMM 2026)
PB  - Atlantis Press
SP  - 1338
EP  - 1347
SN  - 2352-5428
UR  - https://doi.org/10.2991/978-94-6239-801-6_139
DO  - 10.2991/978-94-6239-801-6_139
ID  - Li2026
ER  -