EXPLORATORY TEST ORACLE PREDICTION USING MULTILAYER PERCEPTRON NEURAL NETWORK

Abstract

Software testing is aimed at evaluating a feature, or capability of a program and determine whether it meets its required results. Different software testing approaches have been practiced over the years to improve quality of software. Studies have shown that exploratory testing has become of great importance. The approach gained momentum as it is focused towards error detection in a timely manner and high possibility of testing complex combinations that cannot be designed in test case based techniques. Traditionally, software testing is carried out by focusing on prior test design at the early stages of the software development life cycle, whereas exploratory testing is an approach where the tester does not require to follow a specific pre-designed tests. However, exploratory testing should enable the tester to test the complete system comprehensively. The exploratory testing approach relies on testers’ skills and experience. The main claimed benefits of exploratory testing are the ability of the tester to apply personal knowledge and creativity while testing as well as agility in terms of adapting to changes and effective in working with imperfect documentation. However, no emphasis was put on exploratory test oracle to help the testers. It remained the judgment and responsibility of the tester to determine whether a test executed passed or failed. In general, the exploratory testers designs and executes the tests on the fly and compares the actual output produced by the application under test with the expected output in their mind. Due to the fact that humans are fallible in nature, failures are not always detected by the exploratory testers even when revealed by the test case. This is one of the main drawbacks of relying on human knowledge. Because of the lack of guidelines, there is a high possibility of judging a passed test as failed and vice versa. Humans are more prone to errors when evaluating complex behaviors or detailed specifications and the correctness of the human oracle drops precipitously with an increase in the number of tests to be assessed. Even if it were more reliable, the human oracle is prohibitively very expensive for enormous volumes of test cases.