Building an Intelligent Phishing Email Detection System Using Machine Learning and Feature Engineering
Purna Chandra Rao Chinta, Chethan Moore, Laxmana Murthy Karaka, Manikanth Sakuru, Varun Bodepudi, Srinivasa Rao Maka
Abstract
The prevalence of cybercrime is directly proportional to the growth in the number of people using the internet. There has been evidence of phishing's extensive usage since its beginning, and it is now the most successful cyberattack vector. According to our findings, phishing is the most prevalent kind of cyberattack, and it employs several techniques to deceive its targets. Phishing attacks using malicious URLs, emails, and websites are rather common. Phishing emails continue to pose significant cybersecurity threats, necessitating robust and intelligent detection mechanisms. Using a large-scale phishing email dataset, this research investigates the creation and assessment of sophisticated ML models for detecting phishing emails. Several ML models were used, including CNN, XGBoost, RNN, and SVM. The best answer was suggested by using the BERT-LSTM hybrid model. Featuring an F1-score 99.24, a recall 99.55%, a precision 99.61%, and an accuracy 99.55%, the BERT-LSTM model accomplished remarkable results. Comparative analysis against existing models, including Naïve Bayes, RNN, and SVM, highlighted BERT-LSTM's superior efficacy in detecting phishing emails. Furthermore, training and testing evaluations demonstrated minimal overfitting and consistent generalisation. This study underscores the potential of BERT-LSTM in real-time phishing email detection systems, offering a reliable solution to mitigate phishing threats effectively.