一句话：提出结合量子遗传算法特征选择、量子小波变换与量子-经典分类器集成的混合框架，提升软件缺陷预测性能。
软件缺陷预测（SDP）对早期识别故障模块至关重要，但传统机器学习方法常受无关特征和模型偏差限制。本文提出量子增强混合框架：用量子遗传算法（QGA）选择特征，量子小波变换（QWT）变换特征，再将量子支持向量分类器（QSVC）、量子随机森林（QRF）与经典 SVM 集成为混合分类器，以多数投票融合预测。在 MC1、CM1、KC2、KC3、PC1 五个基准数据集上评估，该集成在准确率、精确率、召回率和 F1 上均优于传统方法，混淆矩阵、ROC 曲线和训练损失收敛图验证了性能提升。该工作展示了量子-经典混合模型用于软件缺陷预测的潜力。
Software defect prediction (SDP) is critical for ensuring software quality by identifying faulty modules early. However, traditional machine learning approaches can be limited by irrelevant features and model bias. In this paper, we propose a quantum-enhanced hybrid machine learning framework for SDP that integrates quantum algorithms for feature optimization with an ensemble of quantum and classical classifiers. We employ a Quantum Genetic Algorithm (QGA) for feature selection and a Quantum Wavelet Transform (QWT) for feature transformation to improve data representation. The selected and transformed features are then classified using a hybrid ensemble comprising a Quantum Support Vector Classifier (QSVC), a Quantum Random Forest (QRF) classifier, and a classical Support Vector Machine (SVM). A majority-voting mechanism combines these models' predictions. We evaluate the approach on five benchmark SDP datasets (MC1, CM1, KC2, KC3, PC1). Experimental results show that our quantum-enhanced ensemble achieves superior accuracy, precision, recall, and F1-score compared to traditional methods. We present confusion matrices, ROC curves, and training loss convergence plots for each dataset, demonstrating consistently improved defect prediction performance. The proposed approach contributes a novel integration of quantum computing into software engineering, highlighting the potential of quantum-classical hybrid models to advance defect prediction.
@inproceedings{MuduliSNP26,
title = {Quantum-Inspired Intelligence for Software Fault Prediction: A Hybrid Ensemble Learning Approach},
author = {Debendra Muduli and Mohammad Shameem and Mahmood Niazi and Sourav Parija},
booktitle = {FSE},
year = {2026},
doi = {10.1145/3803437.3807519}
}