Effective Software Metrics Prediction For Bug Detection Using Machine Learning

Abstract

In the corporate world, with the growth of software business, it is desired to improve the overall software quality. For any software development organization to increase software quality, metrics are a necessary component. Cost-effective test strategy planning and monitoring require the measurement of a software process. Software metrics offer a quantitative method for creating and verifying software process models. Metrics give an organization the information it needs to keep increasing efficiency, cut down on errors, and boost customer acceptability of procedures, goods, and services—all while achieving the intended outcome. This research analyzes the most effective software metrics for bug prediction on various projects NASA datasets: PC1, CM1, KC2 by using machine learning techniques. Most effective metrics contribute significantly to determining bug prediction and help in achieving software reliability. We have performed experimental analysis on different types of performance metrics like precision, recall, F1_score, accuracy and ROC AUC for machine learning models like XGBoost, Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosting Classifier. The feature selection selects important subsets of top 15 metrics among 21 metrics using different ML techniques. The Gradient Boosting feature selection on the PC1 dataset achieved the highest accuracy of 89.8%. The Gradient Boosting feature selection on the CM1 dataset achieved the second highest accuracy of 85%.