Bibliographic citations
Pariona, J., (2017). Clasificación de fuga de clientes en una entidad financiera utilizando el algoritmo Smote para datos desbalanceados en una regresión logística [Tesis, Universidad Nacional Agraria La Molina]. https://hdl.handle.net/20.500.12996/3329
Pariona, J., Clasificación de fuga de clientes en una entidad financiera utilizando el algoritmo Smote para datos desbalanceados en una regresión logística [Tesis]. : Universidad Nacional Agraria La Molina; 2017. https://hdl.handle.net/20.500.12996/3329
@misc{renati/242026,
title = "Clasificación de fuga de clientes en una entidad financiera utilizando el algoritmo Smote para datos desbalanceados en una regresión logística",
author = "Pariona Huarhuachi, Jefferson Clauss",
publisher = "Universidad Nacional Agraria La Molina",
year = "2017"
}
Customer retention has taken much importance in recent years in financial institutions due to aggressive competition from the sector, as well as the autonomy of the client to seek better benefits within all offers that exist in the banking market, which is reflected in the increase in the rate of customers escaped. It has been necessary the implementation of statistical or technical techniques of data mining, in order to build a predictive classifier that can help identify potential customers to abscond. In many cases when classification techniques are applied, it is common to predict class to occur less frequently than other kind: the presence of unbalanced data. I.e. you have fewer customers escaped than not escapees, which represents a drawback since the classifier needs sufficient both kinds of data to be able to learn from them and thus achieve a good prediction. This research proposes the Syntetic Minority Over-sampling algorithm Technique (SMOTE) as a solution to this problem. SMOTE creates instances new starting from a sobre-muestreo of them instances existing, carrying the class minority to a number enough to be considered balanced and the class majority if is necessary reduce it by sub-sampling random. In the present study are validated such benefits with the construction of a model of binary logistic regression with unbalanced data with and without the application of the algorithm of SMOTE; in order to predict the flight of clients in a financial institution. They will be used to measure the precision, the ROC curve and elements of table cross as the specificity and sensitivity testing.
This item is licensed under a Creative Commons License