Software Defect Prediction based on Random Forest Classifier with Artificial Neural Networks
Software is playing an increasingly essential role in many industries. However, defects are not only inconvenient and aggravating, but can also have serious cost for software systems, especially for mission-critical systems. Therefore, software defect prediction models are useful for understanding, evaluating and improving the quality of a software system. Machine learning techniques have been working to make predictions about the defectiveness of software components by exploiting historical data of software components and their defects. In order to predict software defects, many studies using Random forest classifier with Artificial Neural networks (RF-ANN) have been proposed. The effectiveness of our proposed method is evaluated using historical data from the NASA and PROMISE software engineering repository, by comparing it with a k-nearest neighbor, SVM and Random Forest baseline. Our evaluation on a widely used data set shows that our method significantly improves the performance of the proposed classifier.