Heterogeneous Defect Prediction Based on Federated Transfer Learning via Knowledge Distillation

Abstract

Heterogeneous defect prediction (HDP) aims to predict defect-prone software modules in one project using heterogeneous data collected from other projects. There are two characteristics of defect data: data islands, and data privacy. In this paper, we propose a novel Federated Transfer Learning via Knowledge Distillation (FTLKD) approach for HDP, which takes into consideration two characteristics of defect data. Firstly, Shamir sharing technology achieves homomorphic encryption for private data. During subsequent processing and operations, data remains encrypted all the time. Secondly, each participant uses public data to train convolutional neural networks(CNN), the parameters of the pre-trained CNN are transferred to a private model. A small amount of labeled private data fine-tunes the private model. Finally, knowledge distillation realizes the communication between the participants. The average of all softmax output (logits) is used for knowledge distillation to update the private models. Extensive experiments on 9 projects in 3 public databases (NASA, AEEEM and SOFTLAB) show that FTLKD outperforms the related competing methods.

Improved SMOTE and Optimized Siamese Neural Networks for Class Imbalanced Heterogeneous Cross Project Defect Prediction

“…The performance metrics namely accuracy, precision, recall, f-measure, specificity and processing time are evaluated. The state-of-the-art methods of BRFSS [9], CTKCCA [12], KSETE [14], CDAA [15], FSLBDA [17], MSTL-AE [18] and FTLKD-CNN [19] are also implemented in the same simulation setting to compare their performance with the proposed method denoted as GEO-SNN. the other methods with high values of accuracy, precision, recall, f-measure, and specificity and less processing time.…”

“…the other methods with high values of accuracy, precision, recall, f-measure, and specificity and less processing time. The GEO-SNN has approximately 3%, 6%, 3%, 3%, 8%, 6% and 4% high accuracy than BRFSS [9], CTKCCA [12], KSETE [14], CDAA [15], FSLBDA [17], MSTL-AE [18] and FTLKD-CNN [19], respectively. The use of effective class imbalance processing and deep feature learning with parameter tuned SNN has been the major reason for this improvement.…”

“…The results in Table 4 show that the proposed GEO-SNN method has conveniently outperformed the other methods. GEO-SNN has approximately 3%, 8%, 4%, 4%, 10%, 6% and 5% high accuracy than BRFSS [9], CTKCCA [12], KSETE [14], CDAA [15], FSLBDA [17], MSTL-AE [18] and FTLKD-CNN [19], respectively. The proposed GEO-SNN has provided better defect prediction with heterogeneous metrics very effectively.…”

“…The results in Table 5 show that the proposed GEO-SNN method has conveniently outperformed the other methods for the AEEM group source dataset EQ and MORPH group target dataset Velocity-1.4. GEO-SNN has approximately 3%, 9%, 11%, 12%, 10%, 7% and 3% high accuracy than BRFSS [9], CTKCCA [12], KSETE [14], CDAA [15], FSLBDA [17], MSTL-AE [18] and FTLKD-CNN [19], respectively. This performance improvement can be attributed to the parameter tuned deep learning model of SNN using the advanced GEO optimization.…”

“…However, this method also does not consider the class imbalance and the noise problem. Wang et al [19] developed HDP using Federated Transfer Learning based on the knowledge distillation (FTLKD) approach. This approach used trained convolutional neural networks (CNN) to utilize knowledge distillation for detecting the different metrics of projects and improved the accuracy, AUC and G-mean.…”