Knowledge transfer software fault learning

A large amount of work has been done on prediction of software defects and prediction of number of bugs. Most of the approaches are based on training and testing of data related to the same project. But there may not be sufficient data of the project for training. The cross project prediction makes use of data from other projects to predict the defects in a particular project. In this work, we proposed a model for cross domain projects prediction. The training data is take from the projects on one domain and prediction of faults of the other domain software projects is performed. In order to learn from other domain knowledge we have used balancing of the source project data. The balanced data of the source project is used to train the Random Forest classifier on which hyper parameter tuning is applied. To examine the benefits of our approach, we perform experiments on nine projects datasets from the PROMISE repository. Additionally we also compare our approach with the recently proposed cross project prediction: HYDRA model. The results show us good f-measure values obtained by the proposed model for some projects over HYDRA.