Spam Email Detection using Deep Learning Techniques

Anything that is connected to the internet is vulnerable, for example mobile phones, personal laptops, tablets, routers, and smart speakers. Cybercriminals need one point of weakness like unprotected devices or a weak password or any attachment to potentially enter into the system. There is a need to pause before proceeding with any mail or downloading any document or accessing any link in a message because there is a risk of phishing. Every day 320 billion spam emails are sent to many people. According to statistics of spam mail, it was noted that out of every 3000 emails, 1 mail is spam that contains phishing links, malware, fake messages, fake offers, etc. The hacker tries to get confidential information about people, companies, and bank account details. In 2023, Spam mails are still a big real-life problem because some people are still not aware of spam emails and they aren’t able to detect spam mail manually. So, there is a need for the development of a spam detector system that can detect spam emails with higher accuracy. In this paper, there will be a discussion about implementation, execution and obtained results of deep learning algorithms like LSTM (one-directional), BiLSTM (Bi-directional), BERT, and Convolution Neural Networks using a dataset that was downloaded from Kaggle. An accuracy of 98% was obtained with the CNN, 96% was obtained with the LSTM (one-directional) model, 97% with the BiLSTM (Bi-directional) model, and 99% was obtained with the BERT model. The best accuracy ‘of 99%’ with great recall value, less precision, and a great F1 score was attained by implementing the BERT model for spam detection. Keywords-Deep Learning, CIA, SIANN, RFC