ENHANCING SOFTWARE QUALITY THROUGH MACHINE LEARNING DRIVEN DEFECT PREDICTION MODEL

Abstract

Software defects are among the key problems in software engineering as it compromises software reliability, boosts up maintenance expenses and impacts software quality. Traditionally, defects are discovered using a time-consuming technique and, in a large and complex software system, it is not so efficient. This study was interested in software defect prediction application of machine learning techniques to improve software quality assurance in the early stage in this regard. Designing a machine learning model that can predict and detect the defect prone software modules at an early stage of Software Development Life Cycle (SDLC) is the prime objective of research. Historical Software Defect Data sets from PROMISE repository were used in the study. To improve data quality, the following data cleaning, class balancing, data normalization and missing value imputation techniques were employed. The most relevant software metrics were also selected using feature selection techniques like Principal Component Analysis (PCA). Different machine learning algorithms were implemented and tested including Random Forest, Support Vector Machine (SVM) and Naïve Bayes with the performance metrics of accuracy, precision, recall, and F1-score. The outcomes of the experiments showed that the use of machine learning technique had a positive impact on the software defect detection prediction accuracy. Among the models implemented the complex and elaborated models gave better results, with the best model being the Random Forest model. The study shows that machine learning defect prediction models could be beneficial for software developers to reduce the amount of software testing and software maintenance cost, software reliability and software quality as well as at the early stage of software development to identify software components that may be at risk of failure. Moreover, the study emphasized on data pre-processing and feature selection approaches to boost the prediction accuracy and to perform successful software quality assurance procedures.