A Novel Approach for the Detection of Web Service Anti-Patterns Using Word Embedding Techniques

An anti-pattern is defined as a standard but ineffective solution to solve a problem. Anti-patterns in software design make it hard for software maintenance and development by making source code very complicated for understanding. Various studies revealed that the presence of anti-patterns in web services leads to maintenance and evolution-related problems. Identification of anti-patterns at the design level helps in reducing efforts, resources, and costs. This makes the identification of anti-patterns an exciting issue for researchers. This work introduces a novel approach for detecting anti-patterns using text metrics extracted from the Web Service Description Language (WSDL) file. The framework used in this paper builds on the presumption that text metrics extracted at the web service level have been considered as a predictor for anti-patterns. This paper empirically investigates the effectiveness of three feature selection techniques and the original features, three data sampling techniques, the original data, four word embedding techniques, and nine classifier techniques in detecting web service anti-patterns. Data Sampling techniques are employed to counter the class imbalance problem suffered by the data set. The results confirm the predictive ability of text metrics obtained by different word embedding techniques in predicting anti-patterns.