Prediction of Web Service Anti-patterns Using Aggregate Software Metrics and Machine Learning Techniques

Abstract

Service-Oriented Architecture(SOA) can be characterized as an approximately coupled engineering intended to meet the business needs of an association/organization. Service-Based Systems (SBSs) are inclined to continually change to enjoy new client necessities and adjust the execution settings, similar to some other huge and complex frameworks. These changes may lead to the evolution of designs/products with poor Quality of Service (QoS), resulting in the bad practiced solutions, commonly known as Anti-patterns. Anti-patterns makes the evolution and maintenance of the software systems hard and complex. Early identification of modules, classes, or source code regions where anti-patterns are more likely to occur can help in amending and maneuvering testing efforts leading to the improvement of software quality. In this work, we investigate the application of three sampling techniques, three feature selection techniques, and sixteen different classification techniques to develop the models for web service anti-pattern detection. We report the results of an empirical study by evaluating the approach proposed, on a data set of 226 Web Service Description Language(i.e., WSDL)files, a variety of five types of web-service anti-patterns. Experimental results demonstrated that SMOTE is the best performing data sampling techniques. The experimental results also reveal that the model developed by considering Uncorrelated Significant Predictors(SUCP) as the input obtained better performance compared to the model developed by other metrics. Experimental results also show that the Least Square Support Vector Machine with Linear(LSLIN) function has outperformed all other classifier techniques.