Zheng, Xinyu; Ma, Yonghong; Song, Zhigang Source: ACM International Conference Proceeding Series, p 51-56, October 27, 2023, Proceedings of the 2023 4th International Conference on Machine Learning and Computer Application, ICMLCA 2023;

Abstract:

The development of credit scoring models holds significant importance in banking for crediting operations. Combining multiple credit scoring models using a stacking generalization approach often yields improved results. However, selecting the optimal combination of credit scoring models is NP-Hard. In recent years, emerging quantum annealing (QA) techniques have shown promise in the field of combinatorial optimization and are considered potential solutions for NP-Hard problems. Hence, we propose an approach that employs QA for solving the credit scoring model combination optimization problem using a stacking generalization approach. By employing both auxiliary variable substitution and polynomial reduction methods, two distinct Quadratic Unconstrained Binary Optimization (QUBO) models suitable for quantum annealing are formulated. Experimental results demonstrate that the hybrid quantum annealing solver outperforms simulated annealing and simulated quantum annealing in terms of optimization capability across various problem sizes. The polynomial reduction and auxiliary variable substitution methods each exhibit their own advantages in terms of solution quality and computational efficiency.

© 2023 ACM. (13 refs.)