Empirical Assessment of LR- and ANN-Based Fault Prediction Techniques

Abstract

At the present time, because of our reliance on software systems, there is a need for dynamic dependability assessment to ensure that these systems will perform as specified under various conditions. One approach to achieving this is to dynamically assess the modules for software fault predictions. Software fault prediction, as well as inspection and testing, are still the prevalent methods of assuring the quality of software. Software metrics-based approaches to build quality models can predict whether a software module will be fault-prone or not. The application of these models can assist to focus quality improvement efforts on modules that are likely to be faulty during operations, thereby cost-effectively utilizing the software quality testing and enhancement resources. In the present paper, the statistical model, such as logistic regression (LR), and the machine learning approaches, such as artificial neural networks (ANN), have been investigated for predicting fault proneness. We evaluate the two predictor models on three main components: a single data sample, a common evaluation parameter, and cross validations. The study shows that ANN techniques perform better than LR; but that LR, being a simpler technique, is also a good quality indicator technique.