Comparación de modelos predictivos para propensión de compra y su caracterización mediante análisis de datos textuales

Abstract

Recently CRM (customer relationship management) applications and loyalty schemes have become massive, allowing companies to get more information about their customers and monitor their transactions. It is information in addition to showing how is the current customer, you can allow to know in depth and thus can for example mark the most valuable customers, segment the population to create eye-catching products, predict some future behaviors such as desertion, set the next best deal , de ning cross-selling opportunities and customer pro le more prone in a given marketing year. In this article the issue of comparison of predictive models of the possibility of purchase for an individual in commercial campaigns and their characterization is presented by analysis of textual data. predictive methods Neural Networks, Support Vector Machines, Binomial Logistic Regression, Bayesian Model and Classi cation Trees were applied. selecting the best model by cross-validation, your acuraccy, accuracy and area under the ROC curve is performed. In addition, a textual data analysis was performed to relate the opinions of customers and their ability to purchase.