A Triplet Deep Neural Networks Model for Customer Credit Scoring

Abstract

Deep neural networks are widely used in speech recognition and face verification with excellent performance, and they are gradually applied and developed in the field of customer credit scoring. Traditional credit scoring work relies on the two-step modeling process of feature processing and model building, which cannot effectively balance data dimensionality and model performance. Based on this, we put forward a triplet deep neural network model for customer credit scoring. This model makes use of the feature that deep neural networks and metric learning can efficiently extract and utilize data feature information so that two samples with the same label are embedded tightly while two samples with different labels are embedded loosely, so as to improve the accuracy of credit scoring. All experiments are conducted on three customer credit scoring datasets. We select accuracy, precision, recall, f1-score and AUC to evaluate the classification performance of all models. The experiments show that the triplet deep neural networks model can perform customer credit scoring more accurately compared with the now commonly used random forest (RF), deep neural networks (DNN), logistic regression (LR), k-nearest neighbor (KNN) and support vector machine (SVM).