Aplicación de modelos de aprendizaje supervisado para predicción del tipo de contacto de clientes asignados a un BPO de cobranza

Abstract

The paper presents a research exercise on the application and comparison of 3 supervised learning models that correspond to logistic regression, decision trees and neural networks, to predict the type of contact of clients assigned to a collection business process outsourcing; the methodology used is based on the CRISP-DM methodology from which the business understanding, data understanding, data preparation, modeling and results evaluation phases are executed. For which exploratory data analysis is used, through descriptive statistics, data mining with FAMD (factorial analysis of mixed data), in which it is concluded that the decision tree model and neural network with logistic activation function, obtain very similar results in evaluation metrics. On the other hand, logistic regression has the lowest evaluation results compared to the other models, however, the neural networks present less stability in cross-validation compared to logistic regression and the decision tree.