Introducing a vector space model to perform a proactive credit scoring

Many authoritative studies report how in these last years the consumer credit was up year on year, making it necessary to develop instruments able to assist the financial operators in some crucial tasks. The most important of them is to classify the loan applications as reliable or unreliable, on the basis of the customer information at their disposal. Such instruments of credit scoring allow the operators to reduce the financial losses, and for this reason they play a very important role. However, the design of effective credit scoring models is not an easy task, since it must face some problems, first among them the data imbalance in the model training. This problem arises because the number of default cases is usually much smaller than that of the non-default ones and this kind of distribution worsens the effectiveness of the state-of-the-art approaches used to define these models. This paper proposes a novel Linear Dependence Based (LDB) approach able to build a credit scoring model by using only the past non-default cases, overcoming both the imbalanced class distribution and the cold-start issues. It relies on the concept of linear dependence between the vector representations of the past and new loan applications, evaluating it in the context of a matrix. The experiments, performed by using two real-world datasets with a strong unbalanced distribution of data, show that the proposed approach achieves performance closer or better than that of one of the best state-of-the-art approaches of credit scoring such as random forests, even using only past non-default cases.