Software Defect Prediction using Machine Learning

Abstract

A software's most crucial component is its quality. Software Defect Prediction has gained a lot of traction in recent years and has the potential to directly impact quality. Software quality is greatly impacted by defective software modules, which may result in budget overruns, missed deadlines, and significantly increased maintenance costs. There are diverse phases executed to predict the defect in software such as to employ the data for input, pre-process it, extract the attributes and classify the defect. This research work presents numerous algorithms, namely Gaussian naive bayes (GNB), Bernoulli NB, random forest (RF) and multi-layer perceptron (MLP), for predicting the software defect. This work also focuses on developing an ensemble algorithm to enhance the efficacy of predicting the defects. This ensemble consisted of a Principal Component Analysis (PCA) algorithm with class balancing. Diverse parameters such as accuracy, precision and recall are employed for analyzing the results.