Application of Artificial Neural Networks for Assessing the Testability of Object Oriented Software

In this paper, we present the application of neural networks for predicting the software testability using the object oriented design metrics. The testability is generally measured in terms of the effort required for testing. The object oriented design metrics are used as the independent variables and two JUnit based test metrics are used as dependent variables in this study. The software metrics used include different measures concerning size, cohesion, coupling, inheritance, and polymorphism. This study compares the predic tion performance of neural networks to the two types of statistical analysis methods: least squares regression and robust regression. This study is conducted on an agile based software, written in Java having 40K lines of code. The results of the study indicate that the prediction model using neural networks is better than that of the regression models in terms of the statistical measures of the model evaluation. Index Terms — Artificial neural networks; Object oriented; Regression methods; Testability.