Högskolan i Skövde

Phishing remains one of the most persistent and evolving cybersecurity threats, posing significant risks to individuals and organizations. Traditional rule-based phishing detection methods, such as blacklists and heuristic-based approaches, often fail to identify sophisticated phishing attempts, highlighting the need for more adaptive and intelligent solutions. This research explores the effectiveness of advanced Artificial Intelligence (AI) techniques, specifically, Transformer-based Natural Language Processing (NLP) models and Convolutional Neural Networks (CNNs) in improving real-time phishing detection accuracy.

The study develops and evaluates AI-driven models to classify phishing emails and detect fraudulent websites based on textual and visual data. The experimental results demonstrate that Transformer-based NLP models, such as BERT, significantly enhance phishing email detection by analyzing contextual meaning with high precision. Likewise, CNN-based classifiers, including ResNet and EfficientNet, show strong performance in identifying phishing websites through visual analysis. Furthermore, a hybrid approach integrating textual, URL-based, and image-based features achieves superior detection capabilities, outperforming individual models.

Despite the advancements, challenges such as dataset bias, model generalization, and computational complexity must be addressed to improve practical implementation. This research contributes to cybersecurity by providing insights into AI-driven phishing detection methodologies, offering scalable solutions to mitigate evolving threats. By bridging the gap between traditional and AI-based techniques, this study underscores the transformative potential of machine learning in fortifying digital security.