Congestion Prediction in Internet of Things Network using Temporal Convolutional Network: A Centralized Approach

The unprecedented ballooning of network traffic flow, specifically the Internet of Things (IoT) network traffic, has big stress of congestion on today's Internet. Non-recurring network traffic flow may be caused by temporary disruptions, such as packet drop, poor quality of services, delay, etc. Hence, network traffic flow estimation is important in IoT networks to predict congestion. As the data in IoT networks is collected from a large number of diversified devices with unlike formats of data and manifest complex correlations, the generated data is heterogeneous and nonlinear. Conventional machine learning approaches are unable to deal with nonlinear datasets and suffer from the misclassification of real network traffic due to overfitting. Therefore, it also becomes hard for conventional machine learning tools like shallow neural networks to predict congestion accurately. The accuracy of congestion prediction algorithms plays an important role in controlling congestion by regulating the send rate of the source. Various deep learning methods, such as LSTM, CNN, and GRU, are considered in designing network traffic flow predictors, which have shown promising results. This work proposes a novel congestion predictor for IoT that uses TCN. Furthermore, we use the Taguchi method to optimize the TCN model which reduces the number of experiment runs. We compare TCN with the other four deep learning-based models concerning Mean Absolute Error (MAE) and Mean Relative Error (MRE). The experimental results show that the TCN-based deep learning framework achieves improved performance with 95.52% accuracy in predicting network congestion. Further, we design the Home IoT network testbed to capture the real network traffic flows as no standard dataset is available.