Missing imputation algorithms for credit scoring problems

One of important problems in building credit scroring models is missings imputation. Missings in data may either have obvious economic nature or not have one at all. For example missings in data could appear as a result of technical errors in data storage systems. Therefore the need to impute missings in such a way that maximizes the key quality metric for credit scoring models - Gini coefficient. The most interesting approach to restore missing values is based on generative adversarial networks proposed in paper «Gain: Missing data imputation using generative adversarial nets». The main idea is to use latest achievements in training GAN models to build a framework capable of restoring missings in data with high quality in terms of Gini coefficient compared to traditional approaches to missings imputation.