Deep Generative Models for Reject Inference in Credit Scoring

Deep learning for credit scoring: Do or don’t?

“…(2018) suggested a hybrid model where credit records for customers were transferred into a pixel matrix and then used the obtained matrices to construct a convolutional neural network (CNN) in order to predict default. More recently, Mancisidor et al (2019) constructed a deep generative model (DGM) with the goal of improving the classification accuracy of credit scoring models by adding reject applications. However, mainly two deep learning architectures have been previously constructed for application scoring using a standard pre-processing setup, namely a multilayer perceptron network (MLP) and a deep belief network (DBN).…”

Credit Scoring for Micro-Loans

“…To reduce this effect, we are avoiding use of credit score for small fraction of total population. It is worth noting, if this approach is too costly, model based methods can be used instead to account for this bias [8,9].…”