Diseño de una estrategia de recuperación crediticia temprana para clientes Pequeñas Empresas de una institución financiera del Ecuador mediante los algoritmos k-medias y bosques aleatorios.

Abstract

The purpose of this study is to show a strategy for early credit recovery, that is, before the clients reach a definition of “bad clients”. This is possible using modern methodologies based on machine learning algorithms. This model applies to small companies with current credit activity within a financial institution in Ecuador. Clients must be grouped by their intrinsic unique characteristics, focused on general recovery strategies. The algorithm K-means is used for this purpose and should include variables that help to understand the impact of the client within the financial institution. In this case it will be used the probability of default, current amount and remaining term. The algorithm Random Forest calculates the probability of default. The method used for the construction of the model will be implemented in the statistical software R, which is used for the manipulation, processing and graphic visualization of the data.