EFFICACY OF MACHINE LEARNING MODELS IN SOFTWARE QUALITY PREDICTION

Abstract

Software quality estimation is an activity needed at 
various stages of software development. It may be 
used for planning the project`s quality assurance 
practices and for benchmarking. In earlier previous 
studies, two methods (Multiple Criteria Linear 
Programming and Multiple Criteria Quadratic 
Programming) for estimating the quality of software 
had been used. Also, C5.0, SVM and Neutral network 
were experimented with for quality estimation. These 
studies have relatively low accuracies. In this study, 
we aimed to improve estimation accuracy by using 
relevant features of a large dataset. We used a feature 
selection method and correlation matrix for reaching 
higher accuracies. In addition, we have experimented 
with recent methods shown to be successful for other 
prediction tasks. Machine learning algorithms such as 
Xgboost, Random Forest and Decision Tree are 
applied to the data to predict the software quality and 
reveal the relation between the quality and 
development attributes. The experimental results show 
that the quality level of software can be well estimated 
by machine learning algorithms.