Cross-Project Defect Prediction Method Based on Manifold Feature Transformation

Traditional research methods in software defect prediction use part of the data same project to train model and predict label remaining data. However, practical realm development, that needs be predicted is generally a brand new project, there not enough labeled build model; therefore, traditional are no longer applicable. Cross-project uses type similar target model, so as solve problem loss methods. difference distribution between reduces performance prediction. To this problem, paper proposes cross-project method based on manifold feature transformation. This transforms original space into space, then transformed source finally naive Bayes with better performance. A comparative experiment was carried out using Relink dataset AEEEM dataset. The experimental results show compared benchmark several methods, proposed effectively obtains higher F1 value, which an indicator commonly used measure two-class model.