SCALABLE AND EFFICIENT TRAFFIC PREDICTION IN INTERNET OF THINGS (IOT) NETWORKS USING DEEP LEARNING MODELS

Abstract

As the Internet of Things (IoT) is being developed, Internet traffic has been steadily growing, making it difficult to predict and manage traffic. The authors propose a novel scalable traffic prediction model for IoT networks, which is based on deep learning (DL). It particularly focused on the use of advanced DL techniques such as Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks and hybrid solutions that are capable of dealing with the complex and non-linear nature of the IoT traffic. The aim of the suggested models is to capture the temporal and spatial dependency of traffic data, to increase the accuracy and robustness of prediction models. We also introduce an innovative approach that combines the trend and residual parts of the traffic data to achieve more accurate forecasting of the traffic on different time scales. Furthermore, the paper examines the potential difficulties in implementing the model in real time and handling vast data sets, and suggests potential enhancements to increase the model’s efficiency. The prediction accuracy obtained with the real-world IoT traffic datasets used in the experiments is significantly higher than traditional models, as the Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) have been reduced. This work is crucial to the development of scalable, reliable and efficient traffic prediction models for effective traffic management, congestion control and resource allocation for next generation IoT systems.