Adaptive Framework for Predicting Cellular Network Traffic Bursts
Predicting the traffic in cellular networks is becoming increasingly important because of the explosively growing demand and limited availability of resources. The need for prediction of traffic patterns gains ground during seasonal burst or in case of special events. A huge increase in demand of network resources is observed occasionally which otherwise are under-utilized in other time frames. Also, it is noticed that the prediction performance of 5G and LTE networks varies from model to model. The linear model is better for predicting traffic for longer time frames, while the exponential fitting performs well only during shorter time frames. To address these issues in 5G networks, a new adaptive gradient based prediction model (AGBPM) has been proposed. The proposed algorithm (AGBPM) is compared with linear least square support vector machine (LS-SVM) hybridized with particle swarm optimization(LS-SVM-PSO) and ant colony optimization (LS-SVM-ACO) for analyzing the traffic dynamics and predictions. In order to smoothly perform prediction analysis of the proposed AGBPM algorithm, the traffic patterns of various virtual machines have been collected during normal and bursty periods using a tool Wireshark. These traffic patterns are then preprocessed using a newly proposed protocol namely, Adaptive Time Window Protocol (ATWP). Simulation has been done to test the predictive performance of the proposed AGBPM, LS-SVM-PSO and LS-SVM-ACO models using metrics TPR (True Positive Rate), FPR (False Positive Rate), confusion matrix and MSE (Mean Square Error). The results shown in the paper indicate the supremacy of AGBPM.