Avaliação de desempenho dos procedimentos de boosting e bagging em função do critério de classificação por árvore

Abstract

Classification, at its core, is a human activity that categorizes objects based on specific characteristics. In some situations, it is more efficient and practical to performe this task with machines. Historically, statistics has taken different approaches to classification processes, especially more recently with the development of machine learning and pattern recognition techniques, inspired by the functioning of the human brain. Sometimes, given that classification is crucial in many areas, we face challenges such as ineffective models or imbalanced data. In this context, traditional statistical methods may present errors, leading us to search for better solutions. By combining computational and statistical techniques, we can achieve more efficient classifiers. The focus of this work is to study the performance of bagging and boosting procedures using classification by trees whose performances will be evaluated by ROC curve, classification error rate, and cross-validation.