Ensemble one-vs-all learning technique with emphatic & rehearsal training for phishing email classification using psychology

Psychological crackdown of phishing emails can gain major momentum and success through understanding and classifying phishing emails. Phishing is a preferred way of hackers and has always provided back door entry to the file systems through browsers, plugins and malwares. As analysis is getting refined and people can be identified based on several aspects like network traffic and application usage, people with certain contexts can be traced and targeted easily. While hacking system can be specific and requires definite skills and information, phishing emails can be an easy way to target specific people of interest through information and analysis. To counter that, we propose a model to engage people about these vulnerabilities through creating better understanding the contents and psychology of language. To handle the multiclass language classification problem, we propose an ensemble of classifiers with two-stage training. The ensemble method is characterized by data augmented feature training and negative sampling approach-based generalization. Phishing email-based attack is more psychological than technical as it involves the victim voluntarily (through deception) get into it than exploiting technical fallacies and are difficult to segregate. In this work, we have discussed the various psychological aspects in phishing emails from the point of view of natural language processing and learning to detect them.