An Empirical Framework for Web Service Anti-pattern Prediction using Machine Learning Techniques

In todays software industries, the concepts of Web Services are applied to design and develop distributed software system. These distributed software system can be designed and developed by integrating different Web Services provided by different parties. Similar to other software systems, Web Services based system also suffers from bad or poor design i.e., bad design selection, anti-pattern, poor planning etc.. Early prediction of anti-patterns can help developer and tester in fixing design issue and also effectively utilize the resources. The work in this paper empirically investigates and evaluates six classification techniques, 8 feature selection techniques (7 feature ranking techniques and 1 feature subset evaluation technique), and 1 data sampling technique to handle imbalance data in predicting 5 different types of anti-patterns. These all techniques are validated on 226 real-world web-services across several domains. The performance of the developed models using these techniques are evaluated using AUC value. Our analysis reveals that the model developed using these techniques able to predict different anti-patterns using source code metrics. Our analysis also reveals that the best feature selection technique is OneR, data sample is better that without sampling and Random Forest is best classification algorithm for anti-pattern predictions.