Prediction of Quality of Service Parameters Using Aggregate Software Metrics and Machine Learning Techniques

In todays Service-Oriented Architecture (SOA) world, software systems are built by composing web services offered by Service Providers (SPs). There are different SPs offering services for the same set of functional requirements. Service providers are expected to he highly competitive in their offerings to enhance their market. The quality of web services is an important factor that differentiates one service provider from another. Twelve parameters are identified by which quality of service can he measured. The prediction of these twelve QoS parameters help SPs to enhance the quality of their service. Each web service is realized by several programming files. CK and object oriented metrics of the underlying Java files of the web services are important features for predicting QoS parameters of the web service. The aggregated measure, mean, is chosen to be a feature in predicting the QoS parameters in earlier studies. We propose to build prediction models using 16 aggregate measures and show that there is significant difference between these aggregate measures. We find best feature subset using six feature selection techniques and build prediction models using Extreme Learning Machines with different kernels. We show that feature selection techniques might not enhance prediction accuracies and the ensemble algorithm out performs other learning algorithms.