Deep Learning-Based Defect Prediction for Mobile Apps

Abstract

Mobile applications are increasing their popularity every year. However, unrecognized defects within mobile applications can affect businesses due to negative user experience. To avoid this, defects of applications should be reviewed before releases. The well-known methods for defect prevention include Review and Inspection, Walkthrough, Logging and Documentation, and Root Cause Analysis, as well as employing innovative predictive approaches using machine learning. The benefit of these prediction models is that more testing resources can be allocated to fault-prone modules effectively. This study aims to present a defect prediction model for mobile applications. We applied cross-project and used deep learning algorithms including Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), and Long-Short Term Memory (LSTM) to develop a defect prediction model and applied it to Android applications datasets. SMOTE Oversampling technique is used to balance datasets, accuracy metrics such as precision, recall, F1-score, ROC, and AUC to achieve performance, and model results are evaluated with tenfold cross-validation.