Software Quality Prediction: An Investigation Based on Machine Learning

Abstract

Irrespective of the type of software system that is being developed, producing and delivering high-quality software within the specified time and budget is crucial for many software businesses. The software process model has a major impact on the quality of the overall system - the longer a defect remains in the system undetected, the harder it becomes to fix. However, predicting the quality of the software in the early phases would immensely assist developers in software maintenance and quality assurance activities, and to allocate effort and resources more efficiently. This paper presents an evaluation of eight machine learning techniques in the context of reliability and maintainability. Reliability is investigated as the number of defects in a system and the maintainability is analyzed as the number of changes made in the system. Software metrics are direct reflections of various characteristics of software and are used in our study as the major attributes for training the models for both defect and maintainability prediction. Among the eight different techniques we experimented with, Random Forest provided the best results with an AUC of over 0.8 during both defect and maintenance prediction.