Software Defect Prediction using Dimensionality Reduction and Deep Learning

Software Defect Prediction is a significant angle to guarantee programming quality. Deep Learning methods can likewise be utilized for the same. In this paper, we propose to extract a bunch of expressive features from an underlying arrangement of essential change measures using different dimensionality techniques namely PCA and Kernel PCA, and then train a classifier based on the extracted features using Decision tree and Artificial Neural Networks. We utilize five open source datasets from NASA Promise Data Repository to play out this comparative study. For evaluation, we have used three widely used metrics: Accuracy, F1 scores and Areas under Receiver Operating Characteristic curve. It is found that Kernel PCA along with Artificial Neural Network as a classifier outperformed all the other techniques.