Mobile Busy Traffic Forecasting Based On Support Vector Regression Machine

Mobile operators are facing new challenges and competition in network planning,the network capacity expansion and adjustment is the most important. An accurateand scientific traffic prediction can effectively help operators to grasp thedevelopment trend of voice services, reasonable the network planning, so the choiceof the forecasting model is the core issue of traffic forecasting.This paper analyzes the advantages and disadvantages of several trafficprediction algorithms, and according to the characteristics of mobile busy traffic,choose support vector machine model which has good generalization ability to predict.For the parameter selection sensitive problem of support vector machine model,introduced the particle swarm optimization algorithm, and by using the inertia weightand shrinkage factor on its optimization algorithm, in order to avoid the algorithmfalling into the local extreme values, the result of the experiment shows that theproposed model has higher prediction accuracy compared to the basic particle swarmand support vector machine model.In the particle swarm optimization support vector machine model, we only usethe relevant data that is a small amount of labeled sample data to predict. Therefore,for the problem of only use a small amount of labeled samples learning cause learningmachine training inadequate, this paper puts forward a semi-supervised learningalgorithm, introduced the unlabeled sample data and the geometric relationshipbetween the labeled samples and the unlabeled samples into the model. Through theanalysis of several commonly used semi-supervised learning algorithms,semi-supervised learning algorithm based on graph Laplacian deformation isproposed to deformation the kernel function of support vector machine and use theNystrom algorithm to reduce the calculation amount of graph Laplacian. Experimentsshow that semi-supervised support vector machine model has better generalizationability and prediction accuracy compared to support vector machine model.