Design of an application for credit scoring and client suggestion

Abstract

Risk assessment on loan application is vital for many financial institutions. Most financial institutions have already applied methods of credit scoring and risk assessment in order to evaluate their clients in terms. These systems are often based on deterministic or statistical algorithms. In this context, techniques from artificial intelligence and data mining present themselves as valid alternatives to build such classification systems. In this paper some studies are conducted to evaluate the effectiveness of neural networks as a classification system and improvements upon those classifiers are proposed. Furthermore, a suggestion algorithm is also presented to help clients whose loan applications are refused and provide some explanation on why their loan is refused. Finally an agent based architecture is presented to integrate all algorithms presented in this paper.