Efficient Data Preprocessing approach for Imbalanced Data in Email Classification System

Abstract

Email is one of the important means of communication over the Internet. Due to the rapid growth of Internet, usage of email communication for business, personal and other works has resulted into generation of electronic data in exponential order. Applying machine learning techniques on the huge raw data may degrade the performance. Hence, the data has to be prepared for better performance of the machine learning techniques. The preprocessing phase in machine learning applications such as classification, clustering and prediction is intended to reduce the size of data. This paper proposes a new data preprocessing approach for imbalanced data in email classification domain to measure the effects of various preprocessing methods on different machine learning classifiers. Contribution of various preprocessing methods on the imbalanced dataset is discussed. Accuracy analysis reveals that the proposed approach significantly improves the accuracy of all the machine learning classifiers used in this work. The outcome of this work showed that, success rate of logistic regression achieved 90.39% accuracy in the proposed approach.