Automatic Test Data Generation Using Artificial Neural Networks

Abstract

Independent object test is one of the important steps in the object-oriented software test. This kind of test is faced with two problems: firstly, the object which is under test may call methods of other objects and therefore, the independent test of the object becomes impossible. Secondly, the called methods may be time-consuming and as a result the test of the object takes a long time. In order to overcome these problems, a useful method is to use the faked object which simulates the called methods. Faked objects are usually implemented using a table. This table-based implementation results in different problems such as time-consuming table search operation, and more importantly, inability to exact simulation of called methods. Besides, test samples are rare and therefore automatic generation of test samples which span all the code paths within a method has become a challenging problem. In this paper, a new artificial neural net-work-based faked object is proposed which solves the two above-mentioned problems. This paper contains two pro-posed sections: in the first section, the operation of linear functions which are used in programs is simulated. In the second section, the best set of input parameters which are needed to train the artificial neural network of faked object is determined optimally using genetic algorithm. The superiority of the proposed methods is confirmed using different experiments for mathematical, logical and discrete functions.