A credit risk prediction model combining ensemble learning and deep learning through three-way decisions

Credit risk prediction models have long been a focal point of research for financial institutions. Machine learning models use two-way decisions to generate judgement results, which may lead to high decision risks. To reduce the risk of decision errors caused by two-way decisions, we propose a novel credit risk prediction model using three-way decisions and a combination of ensemble learning and deep learning. In this model, ensemble learning methods are employed to evaluate the default probability. An optimization objective is built to learn three-way decision rules and samples that do not require explicit decisions are categorised into boundary regions using three-way decisions. We construct a deep neural network to extract deep features and provide supplementary information for decision-making related to boundary samples. Subsequently, we employ several publicly available datasets and one real-world dataset to conduct experiments. The results demonstrate that the proposed combined model outperforms other benchmarking methods and more effectively identifies defaulters.