SPAMNET: A HYBRID DEEP LEARNING FRAMEWORK FOR ROBUST SPAM EMAIL DETECTION USING MULTI-MODAL FEATURES

Abstract

Spam emails are still making it hard for people to use digital communication systems. They pose security risks and slow down operations. To solve this problem, we are proposing SpamNet, a mixed deep learning model that uses multi-modal feature learning to improve the accuracy of spam identification. SpamNet is different from other methods as it uses a multi-branch design that looks at sequential language patterns, statistical metadata features, and time-based behavioural trends all at the same time. The model is trained and tested on a balanced, publicly available email dataset with 6000 samples that have an equal number of spam and ham emails. The result obtained results during experiments show that SpamNet has an 98.81% validation accuracy, which is a much better than standard deep learning models like DNN, LSTM, and CNN. SpamNet also has lower rates of fake positives and false negatives, which makes it more reliable and stronger in generalisation. The study proves that SpamNet is a useful, small, and scalable approach that can be used in real-time spam filtering systems in a variety of communication settings.