Modelling Interaction Effects by Using Extended WOE Variables with Applications to Credit Scoring

The term credit scoring refers to the application of formal statistical tools support or automate loan-issuing decision-making processes. One most extended methodologies for include fitting logistic regression models by using WOE explanatory variables, which are obtained through discretization original inputs means classification trees. However, this Weight Evidence (WOE)-based methodology encounters some difficulties in order model interactions between variables. In paper, an extension WOE-based is proposed that allows constructing a new kind variable devised capture interaction effects. Particularly, these variables simultaneous pairs single tree. Moreover, can be complemented as usual balance scorecards, enable explaining why individual loans granted not from fitted models. Such explainability loan decisions essential and even more so taking into account recent law developments, e.g., European Union’s GDPR. An extensive computational study shows feasibility approach also enables improvement predicitve capability standard methodology.