Phishing Website Detection Based on Deep Convolutional Neural Network and Random Forest Ensemble Learning

Area of Science:

• Cybersecurity
• Machine Learning
• Web Security

Background:

• Phishing poses a significant global cyber threat, causing substantial financial losses and data breaches annually.
• Existing anti-phishing methods rely on expert-driven feature extraction and third-party services, which are time-consuming and introduce detection delays.
• Limitations of current techniques necessitate more efficient and integrated solutions for robust phishing detection.

Purpose of the Study:

• To propose an integrated phishing website detection method combining convolutional neural networks (CNN) and random forest (RF).
• To develop a system capable of predicting URL legitimacy without accessing web content or utilizing external services.
• To overcome the limitations of expertise requirements and detection delays associated with traditional anti-phishing approaches.

Main Methods:

• URLs are converted into fixed-size matrices using character embedding techniques.
• Convolutional neural networks (CNN) are employed to extract features at various hierarchical levels.
• Multiple random forest (RF) classifiers are used to classify these multi-level features, with a winner-take-all approach for final prediction.

Main Results:

• The proposed integrated model achieved an accuracy rate of 99.35% on the custom dataset.
• On a benchmark dataset, the model attained an accuracy of 99.26%, outperforming existing state-of-the-art models.
• The method demonstrated high efficacy in predicting URL legitimacy without external dependencies.

Conclusions:

• The integrated CNN and RF model offers an effective and efficient solution for phishing website detection.
• This approach significantly enhances detection accuracy and reduces reliance on manual expertise and third-party services.
• The findings highlight the potential of deep learning and ensemble methods in advancing cybersecurity defenses against phishing attacks.