A Study on ML-Based Software Defect Detection for Security Traceability in Smart Healthcare Applications.

Area of Science:

• Computer Science
• Software Engineering

Background:

• Software systems are increasingly complex, raising defect risks.
• Accurate Machine Learning (ML) models for Software Defect Prediction (SDP) are crucial but underexplored.
• This research addresses the need for effective ML techniques in SDP.

Purpose of the Study:

• To investigate and compare various ML techniques for SDP.
• To analyze the impact of Feature Extraction (FE) and Feature Selection (FS) on SDP performance.
• To recommend optimal FE, FS, and ML combinations for accurate defect prediction.

Main Methods:

• Evaluated Feature Extraction (FE) techniques like Principal Component Analysis (PCA) and Partial Least Squares Regression (PLS).
• Assessed Feature Selection (FS) methods including Fisher score, Recursive Feature Elimination (RFE), and Elastic Net.
• Validated ML algorithms (SVM, LR, NB, KNN, MLP, DT) and ensemble methods (Bagging, AdaBoost, XGBoost, RF, Stacking) with FE/FS techniques.

Main Results:

• Feature Extraction and Feature Selection can impact SDP model performance positively or negatively.
• Partial Least Squares Regression (PLS), especially with FS techniques, consistently yielded the most significant improvements.
• Principal Component Analysis (PCA) combined with Elastic-Net showed acceptable performance gains.

Conclusions:

• Feature engineering (FE and FS) is vital for enhancing Software Defect Prediction models.
• PLS-based approaches offer robust and reliable improvements for SDP.
• The study provides valuable insights for selecting appropriate ML and feature engineering techniques for SDP.