Bibliographic citations
Muñoz, E., (2015). Aplicación de redes neuronales y regresión logística para predecir el éxito de la compra de deuda de una entidad financiera [Tesis, Universidad Nacional Agraria La Molina]. https://hdl.handle.net/20.500.12996/1748
Muñoz, E., Aplicación de redes neuronales y regresión logística para predecir el éxito de la compra de deuda de una entidad financiera [Tesis]. : Universidad Nacional Agraria La Molina; 2015. https://hdl.handle.net/20.500.12996/1748
@mastersthesis{renati/242019,
title = "Aplicación de redes neuronales y regresión logística para predecir el éxito de la compra de deuda de una entidad financiera",
author = "Muñoz Muñoz, Emanuel Guillermo",
publisher = "Universidad Nacional Agraria La Molina",
year = "2015"
}
The purpose of this paper is to present and apply the technique of neural networks in order to predict the success of the debt purchase of a financial entity to another. The neural network technique is based on the multilayer perceptron and back propagation algorithm. The use of the technique explains showing different steps: Establishing the structure, the activation function (sigmoid), the learning paradigm, the learning factor, the learning rule, the learning algorithm (back propagation), training and the neural network evaluation. A pre-processing for having data quality, with these clean data was applied the neural networks technique the classification and for evaluating this classification the post processing was performed, one can see that this process is data mining. The highlights in neural networks are the values found for the parameters identified through training. These parameters are called dynamic weights, these weights is the knowledge of the neural network to accomplish the prediction. The classification error obtained by applying to the test data to the neural network trained with the specifications already mentioned was 22.89% and 4.31% for the network of four two neurons in the output layer respectively. When testing with more than seven neurons errors similar or higher ratings were obtained, this was achieved with a greater computational cost. Comparing the results of the same data with logistic regression, these classification errors were obtained 22.89% and 4.31% for the response variable data with four and two kinds respectively. In addition, classification models with the cross-validation technique, in this case 76.57% and 77.29% are percentages prediction efficiency logistic regression and neural networks respectively, only for cases evaluated was obtained where the variable response data was four levels
This item is licensed under a Creative Commons License