Towards machine learning based design pattern recognition

Alhusain,Simon Coupland,Robert John,Maria Kavanagh,Sultan Alhusain

DOI: https://doi.org/10.1109/ukci.2013.6651312

2013-09-01

Abstract:Software design patterns are abstract descriptions of best practice solutions for recurring design problems. The information about which design pattern is implemented where in a software design is very helpful and important for software maintenance and evolution. This information is usually lost due to poor, obsolete or lack of documentation, which raises the importance of automatic recognition techniques. However, their vague and abstract nature allows them to be implemented in various ways, which gives them resistance to be automatically and accurately recognized. This paper presents the first recognition approach to be solely based on machine learning methods. We build a training dataset by using several existing recognition tools and we use feature selection methods to select the input feature vectors. Artificial neural networks are then trained to perform the whole recognition process. Our approach is evaluated by conducting an experiment to recognize six design patterns in an open source application.