A Hybrid Deep Learning Framework for Multi-Class Malicious URL In QR-Code Detection

Abstract

The research deals with the increasing threat of cyber attacks by means of malicious URLs and highlights the inability of conventional detection techniques to counter evolving attack schemes, emphasizing the necessity for sophisticated machine learning techniques. It proposes an innovative Residual Convolutional Neural Network (ResNet)-based architecture for classifying URLs into four distinct categories, namely Benign, Malware, Defacement, and Phishing. Optimized feature selection and extraction mechanisms are incorporated to improve classification accuracy.  
The performance of the model is evaluated based on metrics such as precision, recall, F1 score, and Area Under the Curve (AUC). It is found that the model effectively identifies Malware and Defacement URLs with AUC values of 0.99 and 0.98, respectively.  However, there are challenges associated with the detection of phishing URLs, as indicated by the relatively lower AUC value of 0.92 and F1-score of 0.74, which implies that there are more misclassification errors in this case.
This research has reaffirmed the efficiency of deep learning algorithms in classifying URLs using ResNet while at the same time pointing out areas of improvement in the future, especially with regards to detecting phishing URLs. The research has also been significant to the field of cybersecurity, pointing out the potential of deep learning algorithms in detecting malicious URLs, with future research being proposed to combine ResNet with Recurrent Neural Networks (RNNs) and adversarial learning to combat cyber attacks.