Classification in imbalanced datasets

Abstract

AbstractIn this thesis we study the classification task in the presence of class imbal-anced data. This task arises in many applications when we are interested inthe under-represented (minority) classes. Examples of such applications arerelated to fraud detection, medical diagnosis and monitoring, text categoriza-tion, risk management, information retrieval and filtering. Although there existmany standard approaches to the classification task, most of them have poorgeneralisation performance on the minority class.This thesis studies well-known approaches to the classification problem inthe presence of class imbalanced data, such as Cost-Sensitivity, Bagging for Im-balanced Datasets, MetaCost and SMOTE. The main contribution of the thesisis a new approach to the problem that we call Naive Bayes Sampling. Theapproach is a generative approach. It generates new instances of the minorityclass by bootstrapping values of each feature present in the training data. Ex-periments show the superiority of our approach on 4 UCI datasets and a medicaldataset provided by KULeuven