Examining deep learning’s capability to spot code smells: a systematic literature review

Abstract

Code smells violate software development principles that make the software more prone to errors and changes. Researchers have developed code smell detectors using manual and semi-automatic methods to identify these issues. However, three key challenges have limited the practical use of these detectors: developers’ subjective perceptions of code smells, lack of consensus in the detection process, and difficulty in setting appropriate detection thresholds. While code smell detection using machine learning has progressed significantly, there still appears to be a gap in understanding the effective utilization of deep learning (DL) approaches. This paper aims to review and identify current methods for code smell detection using DL techniques. A systematic literature review is conducted on 35 primary studies from a collection of 8739 publications between 2013 and the present. The analysis reveals that common code smells detected include Feature Envy, God Classes, Long Methods, Complex Classes, and Large Classes. The most popular DL algorithms used are Recurrent Neural Networks (RNN) and Convolutional Neural Networks (CNN), often combined with other techniques for better results. Algorithms that train models on large datasets with fewer independent variables demonstrate exemplary performance. The paper also highlights open issues and provides guidelines for future metric identification and selection research.

Appendices

Appendix A. List of Metrics with their Definitions used in this Study

Acronym	Definition
AMW	Afferent Method and Weighted Method
ATFD	Access to Foreign Data
BUR	Buse Readability Score
CBO	Coupling Between Objects
CDISP	Class Dependency Metric
CINT	Coupling Intensity
CM	Coupling Between Objects
CYCLO	Cyclomatic Complexity
DIT	Depth of Inheritance Tree
FDP	Foreign Data Providers
LAA	Locality of Attribute Accesses
LB	Basic Lack of Cohesion in Methods
LCOM	Lack of Cohesion in Methods
LOC	Lines Of Code
MAXNESTING	Maximum Nesting Level
NOAM	Number of Accessed Members
NOAV	Number of Accessed Variables
NOC	Number of Children
NOM	Number of Methods
NOP	Number of Parents
PAR	Number of Parameters
SDC	Size of Data Context
SEC	Signal Event Coupling
TCC	Tight Class Cohesion
WMC	Weighted Methods Count

Appendix B. List of Primary Studies Selected for this SLR

ID	Title	References
S1	Fusion of deep convolutional and LSTM recurrent neural networks for automated detection of code smells	[39]
S2	Deep Multimodal Architecture for Detection of Long Parameter List and Switch Statements using DistilBERT	[40]
S3	DACOS-A Manually Annotated Dataset of Code Smells	[41]
S4	Improving the Quality of Open Source Software	[42]
S5	Web Service Anti-patterns Detection Using CNN with Varying Sequence Padding Size	[43]
S6	Using Word Embedding and Convolution Neural Network for Bug Triaging by Considering Design Flaws	[44]
S7	Code Smell Detection Using Ensemble Machine Learning Algorithms	[45]
S8	Application of Deep Learning models for Code Smell Prediction	[46]
S9	DeleSmell: Code smell detection based on deep learning and latent semantic analysis	[47]
S10	Feature Envy Detection with Deep Learning and Snapshot Ensemble	[48]
S11	How to improve deep learning for software analytics: (a case study with code smell detection)	[49]
S12	Multi-Label Code Smell Detection with Hybrid Model based on Deep Learning	[50]
S13	Detecting Code Smells with AI: a Prototype Study	[51]
S14	The automation of the detection of large class bad smell by using genetic algorithm and deep learning	[52]
S15	A machine and deep learning analysis among SonarQube rules, product, and process metrics for fault prediction	[53]
S16	Multi-granularity code smell detection using deep learning method based on abstract syntax tree	[54]
S17	An Empirical Study on Predictability of Software Code Smell Using Deep Learning Models	[55]
S18	Two-Pass Technique for Clone Detection and Type Classification Using Tree-Based Convolution Neural Network	[56]
S19	FCCA: Hybrid Code Representation for Functional Clone Detection Using Attention Networks	[57]
S20	MARS: Detecting brain class/method code smell based on metric-attention mechanism and residual network	[58]
S21	Code smell detection by deep direct-learning and transfer-learning	[59]
S22	Deep Learning Based Code Smell Detection	[60]
S23	Feature Envy Detection based on Bi-LSTM with Self-Attention Mechanism	[61]
S24	Deep Representation Learning for Code Smells Detection using Variational Auto-Encoder	[62]
S25	Deep Learning Anti-Patterns from Code Metrics History	[63]
S26	Deep semantic-Based Feature Envy Identification	[64]
S27	Detecting code smells using deep learning	[65]
S28	On the feasibility of transfer-learning code smells using deep learning	[66]
S29	Deep learning based feature envy detection	[67]
S30	A hybrid approach to detect code smells using deep learning	[68]
S31	Cclearner: A deep learning-based clone detection approach	[69]
S32	Finding bad code smells with neural network models	[70]
S33	Deep learning code fragments for code clone detection	[71]
S34	Some code smells have a significant but small effect on faults	[72]
S35	What you like in design use to correct bad-smell	[18]