Ensemble Boosting Algorithms for Software Defect Prediction
Dusa Sai Kiran, Ramesh Ponnala · 2023
In the ever-evolving world of software development, ensuring high-quality software is crucial for meeting user expectations. Software Defect Prediction (SDP) identifies the modules which may contain defects and require substantial testing. Therefore, predicting software defects at an early stage is very much helpful for companies. However, problems like class imbalance and noise in dataset needs to be addressed to build a better model. This paper uses SMOTE for resampling the data and explores the use of ensemble boosting algorithms for predicting software defects. The study utilizes several boosting algorithms, including CatBoost, XGBoost, and LightGBM, and compares their performance on publicly available software defect dataset such as NASA(PROMISE). The proposed ensemble learning model significantly improved the predictive accuracy, outperforming conventional machine learning methods like Logistic Regression and Naïve Bayes. The experimental outcomes show that ensemble boosting model has achieved better results in predicting defects with 86% Accuracy, 0.85 G-mean and 0.92 AUC measures.