Reproduce “Quantum vs. Classical Machine Learning Algorithms for Software Defect Prediction: Challenges and Opportunities” on OpenResearch

About this paper

Md Nadim, Mohammad Hassan, Ashis Kumar Mandal, Chanchal K. Roy

This study conducts a pioneering empirical comparison of Quantum Machine Learning (QML) and Classical Machine Learning (CML) algorithms for software defect prediction. It demonstrates that QML algorithms, particularly the Quantum Support Vector Classifier (QSVC), achieve superior F1 scores and recall in identifying buggy software commits despite facing substantial practical challenges related to current quantum hardware limitations and simulation overhead.

Key insights

• Current QML implementations face significant challenges for SDP, including limitations in available qubits, difficulties with high-dimensional feature spaces (restricted to 15 features), and exponentially long execution times on quantum simulators.
• The Quantum Support Vector Classifier (QSVC) consistently outperformed other QML and classical algorithms in terms of average Recall and F1 scores, achieving the highest F1 score in 13 out of 20 datasets.
• While QSVC excelled in overall balanced performance (F1) and detecting true positives (Recall), classical algorithms like Gradient Boosting, Random Forest, and Support Vector Classifier generally showed higher Precision.

Summary by alphaXiv.