Towards High-Quality Test Suite Generation with ML-Based Boundary Value Analysis
Xiujing Guo, Hiroyuki Okamura, Tadashi Dohi · 2023
In software testing, a protective measure to prevent faults in the code is to ensure that the behavior on the boundary between the sub-domains of the input space is correct. Therefore, designing test cases with boundary value analysis (BVA) can detect more errors and improve test efficiency. This paper presents an ML (machine learning) based approach to automatically generate boundary test cases. Our approach is twofold. First, we train an ML-based discriminator that determines whether a boundary exists between two test inputs. Second, using the outputs of the discriminator, we create test inputs based on Markov Chain Monte Carlo. We conduct experiments to compare the fault detection capabilities of the ML-based approach with concolic testing and manually-performed boundary analysis. Results indicate that the ML-based method outperforms the manually-performed boundary analysis in four of the seven programs tested and concolic testing in three of the seven programs tested.