A Classification Model to Detect Malicious Software in QR Code

Abstract

Quick Response (QR) codes have become very popular in making payments, marketing/authentication, and are frequently used with malicious intent by cybercriminals using them to inject malicious URLs to steal information or install malware. The current study is concerned with malicious QR codes detection with deep learning. A balanced data set of 316,254 benign and 316,254 malicious URLs were turned into an image of QR code to train a Convolutional Neural Network (CNN). On the validation set, the model attained 99.62% accuracy and on the test set, the model attained 100% accuracy, absolute precision, recall and F1 scores. Average processing speed is 61ms per batch which allows real time scanning. CNN was able to outperform other probabilistic models because of its high performance implying that further comparisons can be carried out in future. This work brings to the fore an efficient, scalable solution devised to identify malicious QR codes and enhance the security in the current contactless digital environment.