Automatic detection of Feature Envy and Data Class code smells using machine learning

Milica Skipina, Jelena Slivka, Nikola Luburic, Aleksandar Kovacevic

Abstract: Highlights•We detect Feature Envy and Data Class code smells using pre-trained code embeddings.•We compare handcrafted code metrics with automatically inferred code embeddings.•We test the performance of smell detectors on the large manually labeled dataset.•Embedding capture crucial semantics and do not suffer metric extraction limitations.•Code embedding and code metrics have comparable smell detection performance.