Detecting ham and spam emails using feature union and supervised machine learning models

Abstract

Spam emails are cyber nuisances that cause serious security threats including personal and financial information. Although several spam detection approaches exist, detecting new strains of spam messages is challenging that requires a reliable and efficient intelligent spam email detection approach. This study utilizes features from the text of emails to determine whether it is spam or normal. Multiple features are combined to obtain a higher accuracy for spam email detection. Experiments involve machine learning and deep learning models and the influence of data resampling is also investigated. Performance analysis is done using F1 score, recall, precision, and accuracy, as well as comparison with state-of-the-art approaches. Random forest and logistic regression achieve the highest accuracy scores 0.991 and 0.990, respectively which is much better than existing models.