APPLYING MACHINE LEARNING TECHNIQUES TO PREDICT SOFTWARE QUALITY

Abstract

The prediction of software quality using machine learning (ML) is a rapidly expanding field that employs various ML algorithms to forecast the quality of software systems. Assessing software quality is essential across all phases of development, enabling effective organization of quality assurance practices and facilitating comparisons between projects. Previous studies utilized approaches such as Multiple Criteria Linear Programming and Multiple Criteria Quadratic Programming for quality assessment, alongside algorithms like C5.0, SVM, and neural networks, which achieved only moderate accuracy. This research aims to improve prediction precision by leveraging key features from large datasets through feature selection and correlation matrix analysis. Additionally, we evaluate advanced techniques proven effective in other prediction tasks. Specifically, ML algorithms—including XGBoost (Gradient Boosting), Random Forest, Decision Tree, Logistic Regression, Bagging Classifier, and K-Nearest Neighbors—are applied to predict software quality and reveal relationships between quality metrics and development characteristics. Experimental results demonstrate that these ML methods can accurately forecast software quality levels.