Comparación de Modelos Apilados Bajo los Esquemas de Redes Neuronales y Árboles de Clasificación

Abstract

Neural networks and decision trees have shown to be e cient methods in applications related to classi cation, currently there are still attempts to combine these methods with the purpose of obtaining a more e cient algorithm able to improve the accuracy of its estimates, for this reason this thesis has a purpose of creating and comparing the performance of the neural networks models and decision trees, starting from the combination of di erent decision algorithms through the ensemble method, which was developed by Wolpert in 1992 to increase the predictive accuracy of the models, the idea of this method is to collate the estimation of di erent classi cation algorithms and afterwards develop a nal model from this new dataset. In Wolpert's terminology, the original data and the models constructed for them in the rst step are called models and data collection of level 0, while the data set contains estimates of the level 0 models along with the variable to predict original is denominated as level 1 data, in the same manner the learning algorithm created for this new database will take the name of level 1 model.