Examinando por Autor "Betancur Londoño, Carlos Mario"
Mostrando 1 - 1 de 1
Resultados por página
Opciones de ordenación
- ÍtemModelo de aprendizaje automático para riesgo crediticio de microempresarios regionales según perfil socioeconómico(Fundación Universitaria Los Libertadores. Sede Bogotá., ) Betancur Londoño, Carlos Mario; González Veloza, José John FredyThe credit portfolio is fundamental in a financial entity, therefore, before each credit delivered, the hope is to recover it in times agreed with the client, even so, the risk of non payment during the term of the obligation is latent. The proposal of a prediction model with different techniques that defines the probability of default, can help define the possible socioeconomic causes that imply risk of default. The achievement of the defaults caused at the moment was taken with the objective of identifying clients that could incur in a state of default and risk of non-payment. The modeling was done in order to mitigate or filter the users to whom the credit is granted and helps us define how they can be classified as a potential default holder, this, determined by the profiles provided by the more than 39 thousand individuals that make up the database. The market niche to which the institution is directed is made up of users with limited economic scope to start their business or micro entrepreneurs who require working capital for their ongoing business, all of them with a common interest, to create a business and get ahead with your idea, regardless of academic levels, financial muscle or urban or rural residence. A solid concept of the project and its implementation is necessary. It is essential to be clear about the market niche to which the institution is directed, and for this reason it is important to consider what its profile is. The models exposed in this project have foundations of support for the area of credit studies or central financial evaluation. The modeling procedure was carried out with supervised machine learning methods such as logistic regression, random forest and gradient boosting. Three options of which the random forest was chosen as the best, according to its metrics. The comparison was made with the current credit evaluation methodology and the implications were determined in case of being implemented.