Artificial Intelligent Email Phishing Detection Application

Chen, Jia Yan (2025) Artificial Intelligent Email Phishing Detection Application. Final Year Project (Bachelor), Tunku Abdul Rahman University of Technology and Management.

Text RIS_Chen Jia Yan_Fulltext.pdf Restricted to Registered users only Download (3MB)

Abstract

Email is still a common target for cyberattacks, and the proliferation of phishing messages produced by artificial intelligence has made conventional detection techniques less and less effective. The goal of this project was to develop and deploy an AI-powered email phishing detection application that combines an adaptive feedback loop, fine-grained natural language processing, a bidirectional LSTM model, and secure user authentication. In addition to supporting.eml, PDF, and plain text uploads, the system is divided into user and administration dashboards and offers comprehensive PDF reports in addition to real-time classification results. The development process was incremental, starting with the compilation of the dataset (83 000 emails, including AI-crafted phishing samples), followed by data cleansing, tokenization, and weighted sampling to highlight user-identified corrections. While security measures included multi-factor email OTP, strong password policies, and server-side encryption of session data, model training used cross-validation and hyperparameter tuning. Unit, integration, acceptance, performance, and security assessments were all part of the extensive testing, which confirmed that the classification latency stays below 3 seconds per email, that accuracy reaches 96 percent, and that the false positive/negative rates are significantly lower than those of a baseline classifier. Even on low-end hardware, the lightweight Bi LSTM architecture ensures quick inference by scaling linearly with available CPU or GPU resources. Future work will investigate broader dataset incorporation, speed up real-time processing pipelines, and integrate domain-based threat intelligence to address emerging phishing strategies, even though the system exhibits robust detection, scalability, and extensibility.