Sign in Subscribe
Machine learning

Advancements in Machine Learning for Software Defects Prediction

Strategic Insights

— 2 min read

Investigations into machine learning have recently led to the publication of a new report, focusing on the role of automated software systems in controlling essential operations. The research team from Singidunum University, funded by the Science Fund of the Republic of Serbia and the Intelligent Multi-Agent Control and Optimization applied to Green Buildings and Environmental Monitoring Drone Swarms (ECOSwarm), explored the application of artificial intelligence and natural language processing in enhancing defect identification within source code. The proposed framework combines convolutional neural networks with XGBoost, AdaBoost, and CatBoost classifiers to achieve high accuracy in defect prediction tasks.

Key Takeaways:

  • The study investigates the use of natural language processing and machine learning in analyzing source code to enhance defect detection and prevention in software development.
  • The proposed framework consists of a two-tier approach, with a convolutional neural network handling complex feature spaces and the second tier employing XGBoost, AdaBoost, and CatBoost classifiers for improved defect detection.
  • The accuracy of the convolutional neural network is 80.6% for the defect prediction task, which is enhanced with the second layer to nearly 81.5%.
  • Experiments with publicly accessible datasets show that the NLP approach exhibits superior outcomes, with XGBoost, AdaBoost, and CatBoost achieving accuracies of 99.6%, 99.7%, and 99.8%, respectively.
  • The research suggests that combining artificial intelligence and natural language processing can significantly improve defect identification within source code.
  • The proposed framework has been shown to be effective in software testing domains, exhibiting large potential for defect prediction tasks.
  • The study has been peer-reviewed and published in the journal Neurocomputing, a publication of Elsevier.

Statistics:

  • The accuracy of the convolutional neural network is 80.6% for the defect prediction task.
  • The accuracy of the second tier employ XGBoost, AdaBoost, and CatBoost classifiers is nearly 81.5%.
  • Experiments with publicly accessible datasets show that the NLP approach exhibits superior outcomes, with XGBoost, AdaBoost, and CatBoost achieving accuracies of 99.6%, 99.7%, and 99.8%, respectively.
  • The proposed framework has been shown to be effective in software testing domains, exhibiting large potential for defect prediction tasks.

Sources:

  • VerticalNews, "New Report Prolongs Machine Learning Research"
  • Journal citation: Neurocomputing, 2025; 630
  • Authors:

+ John Philipose Villoth (Singidunum University)

+ Miodrag Zivkovic (Singidunum University)

+ Tamara Zivkovic (Singidunum University)

+ Nebojsa Bacanin (Singidunum University)

+ Mahmoud Abdel-salam (Singidunum University)

+ Mohamed Hammad (Singidunum University)

+ Luka Jovanovic (Singidunum University)

+ Vladimir Simic (Singidunum University)

  • Funding: Science Fund of the Republic of Serbia, Intelligent Multi-Agent Control and Optimization applied to Green Buildings and Environmental Monitoring Drone Swarms (ECOSwarm)
  • Publication: Neurocomputing, a publication of Elsevier
  • Publisher contact: Elsevier, Radarweg 29, 1043 Nx Amsterdam, Netherlands

Read more