VoStackSDD: A Novel Ensemble Technique for Software Defect Density Prediction

Abstract

Software defect density prediction is vital for improving software quality and reducing maintenance costs. Traditional models often fall short in predicting software defect density, whereas our approach focuses on enhancing software defect density prediction. This research paper presents a novel ensemble learning model, VoStack, designed for software defect density prediction. VoStack, a fusion of Voting and Stacking Regressors, is evaluated against several individual machine learning models, including RidgeCV, SVR, Huber, RandomForest, GradientBoosting, and KNeighbors, across nine datasets from the Tera-Promise and GitHub Bug Prediction Repositories. Each model's performance is evaluated through various statistical and error metrics. Results demonstrate that VoStack consistently outperforms individual models, achieving the lowest error rates and highest predictive accuracy across all datasets. Statistical analyses confirm the significance of these performance differences. This study highlights VoStack's effectiveness in enhancing predictive accuracy for defect density prediction, offering a robust approach for software quality assurance.

Keywords