Experimental comparison of ensemble methods with different approaches of Naïve Bayes classifier in mixed data

“Big Data”, named after the massive volume of both structured and unstructured data, led to an explosion of new methods in data manipulation and data analysis. With the growth of Computational statistics techniques to assist these new methods, it is worth to mention that traditional approaches are not replaced but are reinforced to make more robust predictions and give better results. Under this scope, Data Mining is a new science that was recently developed for delivering the extracted information in a way that humans can comprehend and make better decisions. This pool of pattern recognition methods includes Ensemble algorithms, which have been developed over two decades and are very popular in practice due to their ability to boost the predictability performance of statistical methods. These meta­ -algorithms combine several techniques into one predictive model and aim to decrease variance or bias. Under the scope of Pattern recognition where the prediction of the error plays important role, this thesis aims to compare various Ensemble algorithms with a very famous simple probabilistic algorithm, that of Naïve Bayes, which even today is considered as a method with a very low generalization error and can compete the robustness of Ensemble methods.