Cosine Similarity-based Cross-project Defect Prediction

Cross-project defect prediction has been intriguing researchers in terms of metric heterogeneity and new methods are needed in this field. Performing defect prediction through different projects presents valuable information for developers. In this work, a metric matching algorithm namely CSCDP is presented for cross-project defect prediction. The method is then tested on 36 different projects via three classifiers. According to the obtained results, neural network predictor outperforms the others in terms of mean prediction values. Further, selecting training data sets using sparsity analysis creates a favorable effect on testing performance. Last, CSCDP was able to reduce classification error up to 0.65 in Random Forest for F-score