Feature selection approach for improving the accuracy of software bug prediction

Abstract EN

We recently noticed the advancement and growth in the field of artificial intelligence and in its various branches such as Machine Learning (ML) and Deep Learning in various vital fields such as robotics, smart cars, smart cities, health care, software engineering and many other fields.

Software bug prediction are one of the most important ML uses in software engineering.

In addition, the feature selection is one of ML methods that aim to reduce a feature set that are used for building models.

In this paper, we propose to use the Chi-Square feature selection method to calculate features importance, then to build a ML models, first by using top ten important features and second by using top five important features, based on three of well-known ML classifications algorithms, Support Vector Machine, Naïve Bayes and Linear Discriminant Analysis, with adding and exploring more about the effeteness of new metric of code smell intensity, the performance results of our approach against baseline achieved an improvements as average accuracy among nine datasets reaching up to 5.12%, 4.15% and 1% on the NB, SVM and LDA classifiers respectively