Intelligent Software Bug Prediction: An Empirical Approach

Abstract

Despite the fact that, much research has been conducted to improve accuracy in software bug prediction through different Machine Learning (ML) classifiers, not concentrated on the performance evaluation on the applicability of ML algorithms to detect software bugs. This inadequacy is focused on this paper. In this research, we conducted software bug prediction comparison on six different ML algorithms. Moreover, we adopted Code Based Metrics (CBM) to predict software defect through sequential neural network (DL) model and compared it with generic models. The performance of different models has been evaluated and compared based on PROMISE dataset provided by NASA. Results have shown that ML Techniques and DL Approaches have similar bug prediction capabilities where Decision Tree technique performing the worst and Support Vector Machine gave the best results. Also, thoughtful feature selection provides noticeable difference compared to no feature selection during model construction.