移动App即时缺陷预测模型的可解释性方法

在移动App即时缺陷预测领域，大部分研究只关注预测模型的性能，而忽略了模型的可解释性，因此会造成开发人员对缺陷预测模型的预测结果缺乏信任，并阻碍了缺陷预测模型在实践中的应用。主要针对Android移动App即时缺陷预测模型的可解释性展开研究，通过差分进化算法对局部可解释技术LIME方法进行超参优化得到ExplainApp方法，该方法可以对预测结果生成高质量解释。选择了14个实际Android应用程序作为实验对象，最终结果表明，ExplainApp方法可以解释移动App即时缺陷预测模型得到的实例预测结果。ExplainApp方法在拟合优度上要优于原始的LIME方法，可以平均提高94.50%。

Interpretable method of just-in-time defect prediction model for mobile App

In the field of mobile App just-in-time(JIT) defect prediction, most studies only focus on the performance of the prediction model but ignore the interpretability of the model. Therefore, developers will lack trust in the prediction results of the defect prediction model and hinder the application of the defect prediction model in practice. This paper studied the interpretability of the mobile App JIT defect prediction model. And it proposed the method ExplainApp by hyperparameter optimization of the local interpretability technology LIME method via a differential evolution algorithm, which could generate explanations for the prediction results. This study selected 14 real-world Android applications as preposed experimental objects. The final results show that, the method ExplainApp can explain the prediction results of the instance generated by the mobile App JIT defect prediction model. And the method ExplainApp is superior to the original LIME method in terms of the goodness of fit. Specifically, the performance can improve by 94.50% on average.