Email Spam Detection Using Hybrid Model

Email spam as you know; initially looking harmless, those endless get rich quickly scams and false lottery wins could seem benign. Behind the scenes, however, spam is a real pain for both companies and people. It wastes money, clogs inboxes, and more dangerously can slip in phishing attempts or malware. Researchers have tried everything from simple keyword filters to some rather sophisticated artificial intelligence over the years in an attempt to fight it. The truth is, though, spammers are smart. Your filter's degree of predictability will determine how quickly they can evade it. This study helps with this. Rather than depending only on one type of model, such as a CNN or a simple LSTM, we chose to vary things and produce a hybrid deep learning system. CNNs are excellent at identifying small patterns in text, BiLSTMs help understand the whole context of a message (what came before and after), and then, just for that extra punch, we brought in Reinforcement Learning to let the model actually learn from its own mistakes over time. Think of it like assembling a team where each player brings a special ability. Still, we did not stop there, not even near. In the last section of this work and this is the bit I'm most proud of we developed a custom reward-based attention mechanism that changes which parts of an email receive more focus based on whether the model obtained the previous predictions right or wrong. It's sort of like teaching the model to "pay more attention next time," based on past behavior: a bit like how we humans learn following a mistake. I built everything to run on Google Colab with live demos, tested the model on a mix of standard and hybrid datasets, and ensured the architecture is lightweight enough for practical use. So the outcome is Not only does this system know how to adapt but it also catches spam better than many current systems. And that might just be the secret to remain ahead in a world where spam keeps changing its strategies.