Curated Datasets and Feature Analysis for Phishing Email Detection with Machine Learning

Despite continued research, phishing email attacks are on the rise and there is a lack of rich curated datasets for training and testing email filtering techniques. To address this, we produce and release seven curated datasets with 203, 176 email instances for use with machine learning (ML) to distinguish phishing emails from legitimate ones. We create these datasets by meticulously curating phishing and legitimate emails from different repositories. Then to demonstrate that our curated datasets are suitable for the purpose, we conduct a quantitative analysis for evaluating the performance of five ML algorithms. We also analyze the significance and impact of different features within these curated datasets on those ML algorithms. These curated datasets along with the findings from the quantitative analysis will advance the development of a robust defense against phishing attacks.