Research: "Pengembangan Model Deep Learning Berbasis Transformer untuk Mendeteksi Spam Email"
Abstract: In this project, Me and the team developed an automated spam detection model to mitigate cybersecurity risks such as phishing and malware. We evaluated various deep neural network architectures to understand the context and language structure within emails, providing more adaptive protection compared to traditional rule-based methods.
The increase in email usage has been followed by the evolution of increasingly sophisticated spam techniques. Conventional detection methods often fail to recognize dynamic spam patterns. Through a Deep Learning approach, I aimed to create a system capable of deeply extracting textual features to distinguish between legitimate emails (*ham*) and malicious emails (*spam*).
The model development stages conducted by our team include:
The primary focus of this research is to test the model's ability to understand word sequences. Our team analyzed how the **BiLSTM** (Bidirectional Long Short-Term Memory) model can process information from two directions (past and future) to capture sentence context more accurately than traditional unidirectional models.
Experimental results show a significant advantage for the **BiLSTM** model. This model proved to be the most robust in identifying spam emails that have structural similarities to normal emails. Furthermore, integration with CNN layers helps in extracting specific local patterns within the email text.
Automating spam detection using Deep Learning provides high levels of accuracy and resilience. Our team concludes that utilizing language context and local pattern extraction are key factors in building an email security system capable of facing evolving spam techniques in the future.