Ensemble Learning Applications in Software Fault Prediction

Software fault prediction (SFP) plays a crucial role in the software engineering field by allocating testing resources reasonably, reducing testing costs, and adhering to software quality standards. This paper explores the applications of Ensemble learning algorithms and techniques in SFP. The paper presents a three-stage framework that incorporates random forest, bagging, AdaBoost, LogitBoost, Gradient Boosting, XGBoost, and CatBoost individually, as well as an ensemble of all EL algorithms, applying both soft and hard voting combination rules. Additionally, this paper investigates the impact of SMOTE oversampling with ensemble learning algorithms and techniques based on experiments performed on five (JM1, KC, MC2, KC3, and PC5 NASA MDP) cleaned datasets. The performance of the proposed framework was evaluated using metrics such as Accuracy, F1-measure, Matthew’s correlation coefficient (MCC), area under the curve (AUC), and precision. Our findings, consistent with previous research, reveal that the effectiveness of various Ensemble Learning algorithms fluctuates depending on the specific dataset, with random forest often delivering superior performance. Finally, the results also suggest that ensemble learning has a positive effect on SFP models, provided it is carefully orchestrated.