A Combinational QoS-Prediction Approach Based on RBF Neural Network

Quality of Service (QoS) is considered as an important factor to determine the success of a Web Service. Currently, many QoS prediction approaches focus on time series models. However, these approaches only consider linear and nonlinear time series. Analysis of real QoS datasets shows that they are characterized by other behaviors. Incomplete characteristics analysis of existing prediction approaches will result in wrong prediction results. Furthermore, the collected QoS values may miss some data, which will also impact the prediction accuracy. RBF (Radial Basis Function) neural network model can manage the complex linear and nonlinear relationship, with great flexibility and adaptability. Therefore, we propose a novel combinational prediction approach for QoS based on RBF, which chooses the optimal model from the established linear or nonlinear prediction model, and dynamic gray prediction model according to the data characteristics. Next, the predicted results of these models are passed into the RBF training model as the input, and then used for prediction. Using a public QoS dataset and four real-world QoS datasets, we evaluate the proposed approach by comparing it with previous approach. The experimental results show that our approach is better and improves the accuracy and validity.