Implications of semi-supervised learning for design pattern selection

The significant impact of software design patterns on software design quality has led to conducting more research in this field. A design pattern is a proven solution based on software developers’ experience to solve recurring problems, which is used to acquire quality software design. However, due to a large number of design patterns, selecting an appropriate one is quite difficult. To tackle this issue, researchers have proposed different methods to automatically suggest a suitable design pattern (DP) to the designer. Among the various proposed methods, the text classification–based approach has used supervised and unsupervised methods, which have certain issues such as the need for manual dataset labeling, the need for using separate classifiers for each design pattern class, and the multi-class problem. This study addresses the mentioned issues by providing a three-phase method for choosing the appropriate design pattern. The proposed method exploits the semi-supervised learning method. Subsequently, this study proposes an evaluation model using three widely used case studies and 109 real design problems to evaluate the effectiveness of the proposed method. The evaluation results indicate that the performance of the proposed method has improved compared to the supervised learning techniques of Naïve Bayes and KNearestNeighbor.