DDoS Detection in IoT Networks Using Inter Bot Communication by Deep Learning Techniques

DDoS is a powerful attack which prevents normal users from accessing a service (web service as an example), using regular but multiple fake requests. In this article, we studied preventing DDoS attacks from the source side in IoT networks, focusing on the origin of the attack instead of the destination. We rely on the traffic among bots inside the source network to raise a flag and go to alarm mode, which helps improve both detection and prevention accuracy. The key idea is to analyze the packets transferred among the bots (inner traffic) before the actual attack begins. We used the Bot-IoT dataset and separated DDoS and normal records; after purifying the records, we applied classical machine learning methods such as Random Forest, Naive Bayes, SVM, Linear Regression, and decision tree, alongside deep learning methods including LSTM, TCN and GRU-three models commonly used for time series data. Our approach leverages alert mode and attention mechanisms for more intelligent filtering. The results show considerable improvement in DDoS detection and early prevention, and the best results achieved when using TCN method.

کلیدواژه ها:

DDoS ، IoT ، Deep Learning