Neural network with multiple training methods for web service quality of service parameter prediction

Abstract

Web services have several Quality of Service (QoS) properties. Some of the QoS parameters for web services are availability, response time, throughput, modularity, reliability, and interoperability. A client of a web service can have several web services with similar functionality but different QoS properties for application integration. QoS properties play a decisive factor in selecting the best web services from amongst services having similar functionality. Often QoS parameters are not available, not easy to compute or outdated. We present a method to estimate the QoS parameters of web services from the information contained in web service interfaces. We propose a method based on extracting several data, procedural and structural quantity metrics from the web service interfaces and using them as predictors for estimating the QoS properties. We apply neural network method with 6 different training methods for building a predictive model. Our results demonstrate that the proposed approach is effective. Our experimental results reveal that the structural quality metrics outperforms the procedural and data quality metrics in-terms of the RMSE (Root-Mean-Square Error) performance metric. We conclude that the NLM method (Neural Network with Levenberg-Marquardt training method) out performs five other popular neural network training methods.