Web Service Anti-patterns Prediction Using LSTM with Varying Embedding Sizes

Abstract

Anti-pattern includes the concept of wrongdoing and has many options that seem to be right initially; however, it results in hassle in the long run. Studies observe that the anti-patterns in web services make them more susceptible to change proness and fault proness. Anti-patterns can easily lead to error-prone and unmaintainable solutions. This makes it essential to detect anti-patterns at the early stages of software design so that the software developers can restructure the code in the early stages itself. This will save time and effort it would require to address the issues that could stem from anti-patterns in web services. In this work, Sequence classification with LSTM is applied on the WSDL files from the repository to identify four anti-patterns by using sampling techniques. Our findings indicate that LSTM3 performs best out of three classifier models considered in our work with a mean accuracy of 91.75.