Two-step email spam detection: comparing machine and deep learning accuracy

Artificial intelligence (AI) continues to be a transformative field, offering significant contributions to data science by supporting optimal decision-making processes. One notable application of AI is in digital forensics, particularly in spam email classification. This paper presents a two-step approach to differentiate between regular and spam emails. In the first step, emails are evaluated for vulnerabilities based on three key criteria: varying time intervals between Mail Transfer Agents (MTA), the presence of binary attachments, and inconsistencies in IP addresses associated with the same user. In the second step, a comparative study is conducted between Machine Learning (ML) and Deep Learning (DL) algorithms to identify the most effective method for achieving accurate classification results. The findings demonstrate that the Support Vector Machine (SVM) algorithm from ML outperforms the Recurrent Neural Network (RNN) algorithm from DL, achieving an accuracy rate of 96 % compared to 90 %. A notable conclusion from this research is that manual pre-processing leads to more accurate results and better interpretability compared to automatic pre-processing. This highlights the importance of human intervention in certain stages of AI-driven processes, even when using advanced algorithms. The results suggest that a combination of strategic criteria evaluation and algorithm selection is essential for enhancing the precision of spam classification in digital forensics.