Improving Classification Performance for Minority Classes through the use of Positive-Versus-Negative Ensemble Models

摘要

Classification is a commonly used modelling method for data mining. A classification model is a predictive model which is used to predict a categorical value, called a class. Ensemble classification modelling involves the creation of several base models and a combination algorithm for the base model predictions. The classification modelling process uses a set of instances called the training data. Each instance consists of values for the predictor variables and a categorical label called the class. A class is called a minority class if it has a much smaller number of training instances compared to the other classes. This results in a low level of correct classification compared to the classification performance for the majority classes. Positive-versus-negative (pVn) classification has been reported in the literature as an ensemble classification method which is applicable to classification modelling for multi-class prediction tasks. The purpose of this paper is to report on experimental results for the performance of a replication method for improving classification performance for minority classes, using pVn classification modelling. The experimental results demonstrate that the classification accuracy for minority classes can be improved through the use of pVn classification models.