Classificação de dados sensoriais de cafés especiais com resposta multiclasse via Algoritmo Boosting e Bagging

Abstract

Automatic classification methods have been developed in machine learning area in order to facilitate the categorization of data. Among the most successful methods include the Boosting and Bagging. The Bagging works by combining classifiers adjusted in bootstrap samples of the data and the Boosting works by applying sequentially an algorithm to rank the reweighted versions of the set of training data, giving greater weight to the observations misclassified in the previous step. These classifiers are characterized by providing satisfactory results, low computational cost and benefit of implementation simplicity. Given these characteristics, comes an interest in checking the performance of these automated methods compared with traditional existing classification methods in Statistics, Linear Discriminate Analysis and Quadratic. In order to compare these techniques it was used misclassification rates and accuracy of the models. To improve confidence in the use of Boosting and Bagging methods in more complex problems of classification, a study was carried out by applying these techniques in real and simulated data composed of more than two categories in the response variable. In this dissertation, to encourage the implementation of Boosting and Bagging was held in an application Sensory Analysis. We conclude that automatic methods have a good classification performance by providing lower error rates than Discriminant Linear analysis and Quadratic Discriminant analysis in the tested applications.