Comparative Analysis of Feature Selection Methods for Software Bug Classification

Software bug classification is a critical task in software engineering aimed at identifying defects early to improve software quality and reliability. Despite its importance, effectively classifying software defects remains a challenge, necessitating the use of advanced techniques such as feature selection. This research presents a comprehensive study on feature selection methods for software defect classification, to evaluate their effectiveness in enhancing classification accuracy. The study investigates the impact of feature selection methods (filter-based, wrapper-based, and embedding) on model performance using classification algorithms such Naïve Bayes, Support Vector Machines, K-Nearest Neighbour, and Random Forest. The studies are carried out using publicly accessible software defect datasets, and conventional evaluation measures such as accuracy, precision, recall, and F1-score are used to evaluate the effectiveness of each feature selection strategy. The findings of the study confirm the effectiveness of ensemble methods in bug severity classification, with Random Forest achieving notable accuracy rates for different datasets. Additionally, the study highlights the superiority of wrapper feature selection techniques over filter methods, demonstrating their ability to select informative features for defect severity classification. The findings of this work give useful information for practitioners in selecting and implementing feature selection approaches when developing defect classification models, ultimately contributing to the enhancement of software quality and reliability.