Constrained Classification of Large Imbalanced Data by Logistic Regression and Genetic Algorithm

Imbalance in data classification is a frequently discussedproblem that is not well handled by classical classification techniques. Theproblem we tackled was to learn binary classification model from large data withaccuracy constraint for the minority class. We propose a new meta-learningmethod that creates initial models using cost-sensitive learning by logisticregression and uses these models as initial chromosomes for genetic algorithm. Themethod has been successfully tested on a large real-world data set from ourinternet security research. Experiments prove that our method always leads tobetter results than usage of logistic regression or genetic algorithm alone. Moreover,this method produces easily understandable classification model.