Using a neural network to predict test case effectiveness

Test cases based on command language or program language descriptions have been generated automatically for at least two decades. More recently, domain based testing (DBT) was proposed as an alternative method. Automated test data generation decreases test generation time and cost, but we must evaluate its effectiveness. We report on an experiment with a neural network as a classifier to learn about the system under test and to predict the fault exposure capability of newly generated test cases. The network is trained on test case metric input data and fault severity level output parameters. Results show that a neural net can be an effective approach to test case effectiveness prediction. The neural net formalizes and objectively evaluates some of the testing folklore and rules-of-thumb that are system specific and often require many years of testing experience