Cyber attack detection with QR code images using lightweight deep learning models

As information technologies evolve rapidly, servers are being attacked by cyberattacks due to their high values such as cloud, IoT, mobile and desktop applications. Therefore, cyber-attacks have caused great concern in many areas. Although intrusion detection systems play an important role in cyber security, it has become an important data analysis object because it consists of complex system operating data. Traditional intrusion detection systems detect cyber attacks by recording previously detected attacks and comparing them with new attacks or looking for system anomalies. Intrusion detection data is huge, attack types are diverse, and due to the development of hacking skills, traditional detection methods are inefficient. In recent years, many intrusion detection mechanisms, especially machine learning and deep learning, have been proposed to improve traditional intrusion detection technology. In this study, we propose a multi-objective optimization-based hybrid method that enables the use of the most convenient features of light deep learning models in detecting cyber attacks. First, QR code images of bulky data with multiple classes were created. Then, QR code images were trained using MobileNetV2 and ShuffleNet CNN models. Deep CNN models and features of the trained images were extracted, and Harris Hawk Optimization (HHO) algorithm was used to select the most effective features for classification purposes. As a result, as a result of the classification of the selected features with the proposed hybrid model HHO, attack types were detected with an accuracy rate of 95.89%, and it provided superior performance compared to CNN models.