Software Testing Combining the Fusion of Surrogate Model and Evolutionary Algorithm

The primary challenge is the extensive number of iterations required to converge on suitable solutions in using Genetic Algorithms (GA) for test data generation. Each iteration involves executing candidate solutions and evaluating their fitness values, resulting in prolonged time consumption. This limitation can impede the scalability and efficiency of the evolutionary algorithm, particularly in large-scale or time-critical software testing scenarios. Addressing this issue is crucial for enhancing test data generation efficiency. The method we proposed to mitigate this problem is by fusing surrogate models that approximate the fitness values of candidate solutions instead of executing them on the actual software system. The proposed method typically begins with initial candidate solutions (test data), which along with their fitness values, can be used as a training set to develop the surrogate model. Additional samples may be collected during the evolutionary process to enhance the model’s accuracy. Subsequently, the surrogate model is employed to evaluate the fitness values of new candidate solutions. Solutions with high predicted fitness values are then selected as the initial population to execute the program and undergo the genetic operations. The preliminary experimental results show that by utilizing surrogate models, the need for numerous evaluations on the actual system can be significantly reduced, enabling faster convergence and improved efficiency in test data generation.