Credit Scoring via Kernel Matching Pursuit and its Ensemble

Cuimei Zhang, Jianwu Li, Haizhou Wei

Abstract


Credit risk is paid more and more attention by financial institutions,and credit scoring has become an active research topic in computational finance. This paper proposes to applykernel matching pursuit (KMP) and its ensembleto credit scoring. KMP originates from matching pursuit algorithms that append sequentially basic functions from a basis function dictionary to an initial empty basis using a greedy optimization algorithm, to approximate a given function, and obtain the final solution with a linear combination of chosen functions. KMP is the specialmatching pursuit algorithm using a kernel-based dictionary. An outstanding advantage of KMP in solving classification problems is the sparsity of its solution. Furthermore, we also apply KMP ensemble to credit scoring to model the large-scale data set, which is infeasible for the single KMP. Experimental results based on two data sets from UCI repository and one large data set from individual housing loans in a commercial bank of China show the effectiveness and sparsity of KMP and KMP ensemble in building credit scoring model, compared with the classical classification method - support vector machine.


Keywords


Credit scoring; kernel matching pursuit; kernel matching pursuit ensemble; support vector machine

Full Text:

PDF

Refbacks

  • There are currently no refbacks.


Creative Commons License
This work is licensed under a Creative Commons Attribution 3.0 License.

AAEF©: World Science Publisher United States