Hybrid Classification Approach for Software Defect Prediction with Feature Reduction and Clustering

Abstract

Software product refers to the software which is developed for a specific requirement. Simultaneously, engineering deals with the development of product using explicit technical fundamentals and methods. The software defect can be predicted in diverse stages in which data is utilized as input and pre-processed, attributes are extracted, and classification is performed. This research work makes the implementation of several classifiers in order to predict the software defect. These classifiers are GNB (gaussian naive bayes), Bernoulli NB, RF (random forest) and MLP (multilayer perceptron) which are employed with the objective of forecasting the software defect. The performance of the software defect is enhanced by developing an ensemble classifier. In the introduced ensemble classifier, the PCA (Principal Component Analysis) algorithm is integrated with class balancing. Python is executed to implement the introduced model. Diverse metrics are considered to analyze the results concerning accuracy, precision and recall.