A Robust Tuned K-Nearest Neighbours Classifier for Software Defect Prediction

If the software fails to perform its function, serious consequences may result. Software defect prediction is one of the most useful tasks in the Software Development Life Cycle (SDLC) process where it can determine which modules of the software are prone to defects and need to be tested. Owing to its efficiency, machine learning techniques are growing rapidly in software defect prediction. K-Nearest Neighbors (KNN) classifier, a supervised classification technique, has been widely used for this problem. The number of neighbors, which measure by calculating the distance between the new data and its neighbors, has a significant impact on KNN performance. Therefore, the KNN’s classifier will perform better if the k hyperparameters are properly tuned and the independent inputs are rescaled. In order to improve the performance of KNN, this paper aims to presents a robust tuned machine learning approach based on K-Nearest Neighbors classifier for software defect prediction, called Robust-Tuned-KNN(RT-KNN). The RT-KNN aims to address the two abovementioned problems by (1) tuning KNN and finding the optimal value for k in both the training and testing phases that can lead to good prediction results, and (2) using the Robust scaler to rescale the different independent inputs. The experiment results demonstrate that RT-KNN is able to give sufficiently competitive results compared with original KNN and other existing works.©2022 Springer. This is a post-peer-review, pre-copyedit version of an article published in Proceedings of the 2nd International Conference on Emerging Technologies and Intelligent Systems: ICETIS 2022, Volume 2. The final authenticated version is available online at: https://doi.org/10.1007/978-3-031-20429-6fi=vertaisarvioitu|en=peerReviewed