基于DA-SVM的软件缺陷预测模型

摘要

Feature extraction is an important step in software defect prediction technology research. However, the existing feature extraction cannot accurately obtain the nonlinear dependence relations among features, thus these methods are unable to improve the accuracy of software defect prediction model. In this paper, to solve this question we propose a software defect prediction mod-el ( Denoising Autoencoder Support Vector Machine, DA-SVM) which is based on denoising autoencoder and Support Vector Ma-chine. Firstly, the model extracts features by using denoising autoencoder, secondly uses these features as input of support vector machine, lastly, uses this model to predict bugs. Experimental results show that DA-SVM not only improves the accuracy of soft-ware defect prediction model, but also reduces the noise of history data and enhances the robustness of the software defect predic-tion model.%特征提取是软件缺陷预测技术研究中的重要环节,而现有的特征提取方法无法准确获得特征之间的非线性依赖关系,因而无法提高软件缺陷预测的准确性.针对该问题,本文构建基于降噪编码器和支持向量机的软件缺陷预测模型(Denoising Autoencoder Support Vector Machine,DA-SVM).首先利用降噪编码器进行特征提取,然后将提取的特征作为支持向量机的输入向量,最后再进行软件缺陷预测.实验结果表明,DA-SVM提高了软件缺陷预测的准确度,同时降低了历史数据中的噪声,增强了软件预测模型的鲁棒性.