BUGOPTIMIZE: Bugs dataset Optimization with Majority Vote Cluster-Based Fine-Tuned Feature Selection for Scalable Handling

Abstract

Software bugs are prevalent in the software development lifecycle, posing challenges to developers in ensuring product quality and reliability. Accurate prediction of bug counts can significantly aid in resource allocation and prioritization of bug-fixing efforts. However, the vast number of attributes in bug datasets often requires effective feature selection techniques to enhance prediction accuracy and scalability. Existing feature selection methods, though diverse, suffer from limitations such as suboptimal feature subsets and lack of scalability. This paper proposes BUGOPTIMIZE, a novel algorithm tailored to address these challenges. BUGOPTIMIZE innovatively integrates majority voting cluster-based fine-tuned feature selection to optimize bug datasets for scalable handling and accurate prediction. The algorithm initiates by clustering the dataset using K-means, EM, and Hierarchical clustering algorithms and performs majority voting to assign data points to final clusters. It then employs filter-based, wrapper-based, and embedded feature selection techniques within each cluster to identify common features. Additionally, feature selection is applied to the entire dataset to extract another set of common features. These selected features are combined to form the final best feature set. Experimental results demonstrate the efficacy of BUGOPTIMIZE compared to existing feature selection methods, reducing MAE and RMSE in Linear Regression (MAE: 0.2668 to 0.2609, RMSE: 0.3251 to 0.308) and Random Forest (MAE: 0.1626 to 0.1341, RMSE: 0.2363 to 0.224), highlighting its significant contribution to bug dataset optimization and prediction accuracy in software development while addressing feature selection limitations. By mitigating the disadvantages of current approaches and introducing a comprehensive and scalable solution, BUGOPTIMIZE presents a significant advancement in bug dataset optimization and prediction accuracy in software development environments.