Automated regression testing of computational algorithms based on neural networks

Abstract

Software testing phase in our days became a very important part of software development process. Automation testing plays a key role in agile projects. To reduce the time and effort spent for testing, different extended approaches are used. In our days the artificial intelligence is used for providing different algorithms in different spheres of IT. This article is about the combining the usage of testing artifacts and testing approaches with neural networks working in pair to classify the testing oracle results. Our algorithm is based on QA test case testing concepts, which is laid to neural network usage approach. Test cases are used for verifying whether a tested functionality satisfy the requirements. We created a set of rules how to execute and optimize needed data from test cases to teach the separate neural networks for simulating the tested functionality behavior. According to this, the tested application is somehow substituted by the learning network, which is used for examining the future functionality releases, influenced by hidden errors during regression. The difference of network and application results is combined by the specially created comparison tool calculations that are the basis of created testing oracle. Also we described the approach for creating optimized neural network structure for learning algorithm, that is back propagation in our case