Comparative Analysis of Software Defect Prediction Using Dimensionality Reduction

Abstract

Software defect prediction (SDP) is a dynamic research issue in the field of software development life cycle. It is very helpful in the testing phase of the life cycle of software development. It helps ensure the quality of the software being generated. In this particular paper, we have done a comparative analysis of various dimensionality reduction techniques such as principal component analysis (PCA), kernel PCA, incremental PCA, and sparse PCA, with random forests (RF) and artificial neural networks (ANN) as classifiers. We have collected the data from the android git repository and extracted metrics from 2 different versions of android. For comparing results, we have used three different metrics for a total of 10 experiments run on a dataset combined from 2 different versions of android. The result metrics we used are F1 score, area under receiver operating curve, and accuracy.