查看文章

Accepted: 12/May/2019, Published: 31/May/2019 Abstract—Code smells identified by Fowler [1] is as symptoms of possible code or design problems. Code smells have adverse affecting the quality of the software system by making software challenging to understand and consequently increasing the efforts to maintenance and evolution. The detection of code smells is the way to improve software quality by recovering code smells and perform the refactoring processes. In this paper, we propose a code-smells detection approach based on a decision tree algorithm and software metrics. The datasets we used to train the models are built by reforming the datasets used by Arcelli Fontana et al. work [2]. We use two feature selection methods based on a genetic algorithm to select the most essential features in each dataset. Moreover, we use the grid search algorithm to tuning the decision tree hyperparameters. We extract a set of detection conditions using decision tree models, that are considered as prediction rules to detect each code smell in our binary-class datasets.