Accurate shellcode recognition from network traffic data using artificial neural nets

This paper presents an approach to shellcode recognition directly from network traffic data using a multi-layer perceptron with back-propagation learning algorithm. Using raw network data composed of a mixture of shellcode, image files, and DLL-Dynamic Link Library files, our proposed design was able to classify the three types of data with high accuracy and high precision with neither false positives nor false negatives. The proposed method comprises simple and fast pre-processing of raw data of a fixed length for each network data package and yields perfect results with 100% accuracy for the three data types considered. The research is significant in the context of network security and intrusion detection systems. Work is under way for real time recognition and fine-tuning the differentiation between various shellcodes.