Feature selection, which is an indispensable component in the credit scoring, focuses on reducing irrelevant data and simultaneous improving model performance and interpret-ability, as to expand the authenticity and reliability of credit score. Recent studies suggest that multi-objective framework, which are more comprehensive, objective and relevant than single-objective decision-making, can improve the quality of credit scoring models. We follow this suggestion and present a two-stage multi-objective feature selection approach which considers multiple filter indicators and performance metrics in the filter stage and wrapper stage, respectively. We use Data Envelopment Analysis (DEA) is employed to address the multi-objective decision-making problem from the perspective of Pareto efficient frontier, and a common machine learning technique—Support Vector Machine (SVM) to conduct empirical experiments on eight credit scoring data sets, the results demonstrate that the two-stage multi-objective feature approach outperforms several existed feature selection techniques.


Paper Number 1604; Track Blockchain; Short Paper


When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.