| Abstrakt: |
Code smells (CS) are severe violations of software development fundamentals that degrade source code quality. Several CS types are involved in the program code which has an immediate influence on future maintenance tasks and code changing activities. It is an essential part to find CS automatically by digging into the underlying causes and fixes them properly through refactoring. As a result, CS must be detected automatically. The Deep learning (DL) based approach was proposed to detect CSs by utilizing various features for various smells; however, this DL would perform effectively to detect the single class CS only. In general, DL algorithms may learn beneficial CS detection characteristics and fine tune training data which could be used for various classes of CS detection together. Hence, an advanced DL Based Multiple Class type CS detection (DLMCCMD) is proposed in this article to detect multiple CS categories like large class, misplaced class, lazy class and data clumps automatically. The features of these source codes are extracted from semantic relationship between identifiers in the program code. The CNN-LSTM structure is developed to classify the selected feature which contains source code information and code metrics. The obtained information is merged for positive testing of source code programs with less computational time. [ABSTRACT FROM AUTHOR] |
