HomeNet Shield

Abstract

The aim of our project is to detect intrusions in an IoT network by using deep learning algorithms to analyze various aspects of network traffic, such as packet flow rate, flow duration, packet volume, and so on. The CICIDS2018 dataset will be used since it contains attributes that are relevant to an IoT environment.  The goal is to employ a hybrid network to analyze and classify a tuple as benign or malicious, such as a Botnet or DDoS attack. All the rows with NaN data must be removed from the data set. The types of attacks are classified using a single hot encoding. The Cu-DNNGRU and Cu-DNNBLSTM neural networks combine to form the hybrid network. Both networks' Cuda variants are employed because they have been demonstrated to be nearly 5x quicker than non-cuda variants. Both networks assist in classifying data points in a dataset as benign or malicious, and subsequently assigning a Botnet or DDoS class label. The goal is to catch such invasions in the act.