Machine Learning Algorithms for Delay Prediction in IoT and Tactile Internet

Within the domain of fifth-generation (5G) networks, which represent the next generation of cellular systems, the combination of state-of-the-art artificial intelligence (AI) with revolutionary concepts like globally networked products and touch-based internet is envisaged. The incorporation of artificial intelligence technology into 5G networks presents an opportunity to provide innovative services with significant and long-lasting effects. In particular, network traffic prediction—a critical component of guaranteeing the security, dependability, and quality availability in data networks—is greatly aided by machine learning (ML) technologies, which are skilled at identifying nonlinear patterns in the environment. This research uses a forward-looking k-step-ahead prediction approach to estimate delays in the Internet of Things (IoT) and tactile internet using a nonlinear autoregressive model with external data (NARX)-enabled recurrent neural network (RNN). The machine learning training process uses four different training functions: Levenberg–Marquardt backpropagation (Trainlm), conjugate slope backpropagation with Fletcher-Reeves updates (Traincgf), robust backpropagation (Trainrp), and Bayesian regularization backpropagation (Trainbr). Mean squared error, root average square error, and mean absolute percentage of error (MAPE) functions are used to analyze the accuracy of projected delays, providing a thorough evaluation of the forecast performance.