Hybrid Stacked MemoryNet–CNN with Reciprocal Points Learning for Open-Set Recognition in Unknown DDoS Attack Detectio

Abstract

In the ever-evolving landscape of cybersecurity threats, we can see that Distributed Denial of Service (DDoS) attacks remain as a challenging attack to detect, and particularly when the type of attack is unknown or if the attack is previously unseen at such times this intrusion attack detection is becoming difficult. There are many algorithms such as machine learning and deep learning algorithms which gives highly effective results in identifying known attack patterns, but these algorithms often fail to identify and generalize unknown or any novel attacks effectively.  a novel Open-Set Recognition mechanism is integrated with Reciprocal Points Learning (RPL) and also integrated with a 1D Convolutional Neural Network (CNN1D), they used this mechanism on both known and unknown DDoS attacks, this mechanism contains OSR layer which is designed to tell the difference between known and unknown DDoS attacks, whereas the integrated Reciprocal Points Learning (RPL) will apply distance based reasoning, this distance based reasoning will determine the similarity between input features and known class centers which are in the feature space. But this model was giving accuracy of 93% so, for further improvement of accuracy we have proposed a Hybrid Stacked Memory-Net + CNN-RPL model. This proposed model is trained by using CICIDS2017 Wednesday dataset (known attacks) and validates with the Friday dataset (unknown attacks). Our proposed Hybrid Stacked Memory-Net + CNN-RPL model gave 99.63% accuracy. our model presents a robust solution to modern intrusion detection.