Classifying Internet of Things (IoT) devices and non-IoT devices in data center edge network using statistical and network flow features by one dimensional convolutional neural network

The development of edge computing has led to the aggregation of traffic from IoT and non-IoT devices at edge data center networks located closer to cloud and computing resources. To provide application-specific Quality of Service (QoS) for these devices, classification of IoT and non-IoT devices at edge data centers is crucial. Our research suggests using a simple, compact, and computationally less expensive shallow one-dimensional convolutional neural network (1D-CNN) for real-time processing needs. The model was trained on a dataset consisting of 22 IoT devices and 7 non-IoT devices. The network flow and statistical features extracted from the network traces of these devices were fed into the model. The research demonstrated that with just 15 features, the model can achieve around 94% accuracy, and with 55 features, nearly 97% accuracy in classifying IoT and non-IoT devices.