Network Traffic Prediction and Congestion Control Using a Feed Forward Neural Network

The surge in Internet of Things (IoT) usage has precipitated the need to replace classical internet network prediction and congestion control methods with a more efficient and reliable learning-based approach. Many machine-learning based approaches have been offered by different studies to solve this problem but not without limitations. In this study, we propose a new feed forward neural network with a multi-channel routing congestion control system (FFNN-MCRCCS) to optimize network prediction and congestion control problem. The nodes in the proposed network are mutually connected to enable the firing of neurons in a more decentralized manner to update their weights in order to call the recursive square function to predict the traffic. The exponent value of the recursive function serves as a parameter to determine the channel with the minimum congestion threshold for packet transmission. The results of the experiment show that the proposed model has the capability to optimize network traffic for effective utilization of network resources more than the baseline and the counterpart methods. The performance evaluation metrics used for the experiment include mean square error throughput, link utilization, average jitter, settling time, queue length and network and network convergence.