Utilizing a Neural Network for Software Testing

Author Name : Nidhi Sharma, Pranita Singh, Durga Prasad Roy, Vikash Sawan, Saurabh Bhardwaj

ABSTRACT

The life cycle of software development is not complete without software testing. A test "oracle" is required to decide whether or not a given test case reveals a flaw because the goal of testing is to ensure that an application complies with its specification. The actual cost of the testing process and the associated maintenance costs can be decreased by using an automated oracle to support the actions of human testers. In this article, we provide a novel idea: utilising an artificial neural network as a tested software system's automated oracle. The backpropagation approach uses a series of test cases applied to the system's original version to train a neural network. The "black-box" technique serves as the foundation for network training. since the algorithm is only given the system's inputs and outputs. The trained network can be used as an artificial oracle to assess the accuracy of the output generated by fresh software that might contain bugs. Here, we show the results of an experiment that used a two-layer neural network to find errors in a tiny credit approval application's modified code. The outcomes seem encouraging for a variety of implanted defects..

Keywords: Automated software testing, Mutation testing, Black-box testing, Artificial neural networks