Enhancing Software Fault Prediction with Machine Learning: A Comparative Study on the PC1 Dataset
Sai Krishna Gunda
Abstract
Software makes the world go faster in the modern world, software is everywhere and vital to economy (and environment, education etc.) The dependency on these technology has proliferated, and software systems have become more complex. Traditional statistical approaches like Logistic Regression and Decision Tree have been utilized for RTL-based software defect prediction but they are outdated regarding the nuances of recent research demands. This work investigates whether machine learning approaches can help to advance software defect prediction by employing the PC1 dataset from PROMISE repository. The performance of Logistic regression and Decision tree is compared on MAPE, explained variance, Precision G measure F-measure AUC Accuracy. The results that can be seen in the above ROC charts suggest Logistic Regression does better on AUC and accuracy, but Decision Tree has a more balanced trade off between Recall an Precision. This research contributes to the field of software reliability engineering by applying state-fo-the-art machine learning methods in order to provide new method and insights toward improved accuracy and precision for detection / prediction of software faults.