A Scalable Hybrid Ensemble Model for Text Classification

Text classification is a major problem and an evolving area of wide research with many algorithms had been already proposed. With the recent advancements in the field of Ensemble Learning there are many new techniques consistently emerging that need to be implemented for text classification in search of a better classifier. In this paper, a machine learning model BagBoo, which is a combination of Bag + Boo, where Bag and Boo are representing the Bagging and Boosting respectively. BagBoo model in essence gains its performance from using Bagged ensemble of Boosted trees. In this paper, it is a specific model we use to build our own text classifier using the already existing or some self-modified Bagging and Boosting techniques. We run evaluations on Reuters 21578 data sets 10 most frequent classes, SMS Spam Collection [1] data set, DB World e-mails Data Set and shown through results whether our implementation of BagBoo gives a better performance for text based classification. Results show that the proposed method can achieve an accuracy of 78.94% and 94.21% with SVM and J48 classifier on DBword datasets.