Construction of error prone module reusing prediction model

在易出错模块预测中，我们使用一些预测算法和一些训练数据构造一个预测器，然后利用该预测器预测测试数据的错误倾向，因此为每个测试数据构建合适的预测器是一个重要的问题。一开始我们尝试根据训练数据和测试数据的相似度进行数据调整，但我们发现其他项目特有的特征会影响预测的准确性。其次，我们考虑了基于知识的预测器挖掘方法。该方法利用多种预测算法和项目具体数据，构建由多个项目数据及其预测精度数据组成的知识库。我们构建了测试数据集，并将这些数据集与公共数据集进行对比，验证了该方法的有效性。我们只展示了这种方法的可能性，因此需要进一步的考虑和原型预测工具的构建。

In error prone module prediction, we construct a predictor using some prediction algorithm and some training data, then we predict the error proneness of testing data by using this predictor, so construction of adequate predictor for each testing data is an important problem. At first we tried data adjustment based on the similarity between the training data and the testing data, but we found that other project specific characteristics affect the accuracy of prediction.In the next place, we considered knowledge based predictor mining method. In this method we construct the knowledge base which consists of many project data and their prediction accuracy data using several prediction algorithms and the project specific data. We constructed test data set and by using these data set and public data set, we showed the effectiveness of this method.We showed only the possibility of this method, so further consideration and prototype prediction tool construction are needed.