A survey on the application of deep learning for code injection detection

Code injection is one of the top cyber security attack vectors in the modern world. To overcome the limitations of conventional signature-based detection techniques, and to complement them when appropriate, multiple machine learning approaches have been proposed. While analysing these approaches, the surveys focus predominantly on the general intrusion detection, which can be further applied to specific vulnerabilities. In addition, among the machine learning steps, data preprocessing, being highly critical in the data analysis process, appears to be the least researched in the context of Network Intrusion Detection, namely in code injection. The goal of this survey is to fill in the gap through analysing and classifying the existing machine learning techniques applied to the code injection attack detection, with special attention to Deep Learning. Our analysis reveals that the way the input data is preprocessed considerably impacts the performance and attack detection rate. The proposed full preprocessing cycle demonstrates how various machine-learning-based approaches for detection of code injection attacks take advantage of different input data preprocessing techniques. The most used machine learning methods and preprocessing stages have been also identified.