Abstract:
"The use of machine learning (ML) techniques for predicting software defects is the main goal of
this study. By addressing challenges including redundancy, correlation, and feature irrelevance,
ensemble learning assists ML models perform better. An in-depth analysis of Software Defect
Prediction (SDP) employing ensemble approaches and machine learning techniques with
explainable artificial intelligence (XAI) is presented in this study. It investigates the use of
ensemble approaches, such as stacking, to combine the capabilities of various base models,
including Random Forest, Naïve Bayes, SVM, Logistic Regression, XGBoost, AdaBoost, and
Decision Trees, in order to improve the accuracy of defect prediction models while employing a
variety of preprocessing techniques including SMOTE. Stacking ensemble proved to be the best
model while achieving 80% of predictive accuracy. Among individual model training, Random
Forest performed the best achieving 79% of predictive accuracy. Logistic Regression achieved the
lowest predictive accuracy with 65%.
It underlines the need of XAI techniques in Software Defect Prediction in addition to model
creation. By utilizing SHAP values and LIME, it offers insightful explanations for model
predictions and useful insights into the elements causing software defects. By bringing
transparency to complicated black-box models, these XAI approaches assist stakeholders and
software developers better understand and utilize the defect prediction process.
Overall, this study makes a significant contribution to the field of software defect prediction by
emphasizing the role of ensemble approaches and XAI in improving predictive insights and
providing useful information for software development. Combining modern machine learning
techniques with understandable explanations creates new opportunities for accurate and useful
defect prediction, which ultimately helps software development practices and the production of
high-quality software products."
