Android Smells Detection Using ML Algorithms with Static Code Metrics

Mobile applications development rate is predominantly increasing in comparison with the regular applications. These mobile applications prove to be change frequently according to the user requirements. Moreover, these changes in the code base may introduce some bad design practices that are called as bad smells, which can lead to a higher maintenance cost and degrade quality of the software. A very less attention has been given in the detection of code smells in the mobile applications that are also called as android smells. This research contains the rules in combination of software metrics and their threshold values to detect the bad smells in the android applications. The proposed rules are computed using three different machine learning algorithms. This framework has been applied to 2896 instances of the android applications which are open-sourced on GitHub. The android code smells MIM, LIC, DTWC and SL have been considered for the generation of detection rules and are validated using 10-fold cross validation method. The machine learning algorithm JRip furnished the best result for the android smells up to 90% overall precision, which is quite sufficient to justify the results.