Temporal pattern based QoS prediction

Abstract

Quality-of-Service (QoS) is critical for selecting the optimal Web service from a set of functionally equivalent service candidates. Since QoS performance of Web services are unfixed and highly related to the service status and network environments which are variable against time, it is critical to obtain the missing QoS values of candidate services at given time intervals. In this paper, we propose a temporal pattern based QoS prediction approach to address this challenge. Clustering approach is utilized to find the temporal patterns based on services QoS curves over time series, and polynomial fitting function is employed to predict the missing QoS values at given time intervals. Furthermore, a data smoothing process is employed to improve prediction accuracy. Comprehensive experiments based on a real world QoS dataset demonstrate the effectiveness of the proposed prediction approach.