In view of the problems of low transparency and non-interpretability caused by the widespread use of black-box bigdata credit evaluation techniques such as integrated learning and deep learning in credit investigation, a credit evaluation model interpretation method based on propensity score was proposed. The general framework can be used to conduct explanatory analysis on the black box model of big data credit investigation, so as to make it meet the KYC and KYB requirements in the financial field, and improve the applicability of machine learning, deep learning and other algorithms in the field of credit investigation.