Forecasting Research on the Wireless Mesh Network Throughput Based on the Support Vector Machine

Abstract

Parameters such as network busy rate, number of nodes as well as packet size that affected the wireless mesh network (WMN) throughput were selected as the driving factors which restricted the WMN throughput. A WMN throughput prediction model has been developed based on machine learning methods and experimental study to predict the throughput of IEEE 802.11 WMN. Three kernel functions have been testified and compared through MATLAB. The radial basis kernel function was selected as the support vector regression (SVR) kernel function predication model and its parameters were decided by K-fold cross validation (K-CV) and grid search methods. The proposed prediction model was validated by the data which was simulated in network simulator (NS2). In addition, a prediction model of Mesh network throughput has been established based on back propagation neural network (BPNN). The performance of the models were evaluated using the mean square error and mean absolute error. The experimental results show that the prediction precision of the proposed SVR-based model is a little bit higher than that of the BPNN model. The establishment of the WMN throughput prediction models provides the basis for building, managing and planning rational network structures.