Presentation Information
[A-10-08]Interpretable Boundary Condition Estimation for Software Testing Based on Input-Output Distributions
〇Kazuki Namba1, Satoshi Katsunuma1, Shota Kimata1, Kazuma Hashimoto1, Tatsuhiko Miyata1 (1. Hitachi, Ltd.)
Keywords:
Large Language Model,Black-box Testing,Boundary Condition Estimation,Interpretability
In recent years, large language models (LLMs) have been expected to improve the efficiency of software testing. In black-box testing, the evaluation of test case coverage and validity depends on individual engineers, leading to variability in review quality. Although test case generation based on partitioning the input–output space is effective, boundary conditions are often expressed only as numerical thresholds, requiring specifications or models of the software to interpret their meaning.This paper proposes a method for estimating interpretable boundary conditions based on the distribution of input–output data. The proposed method consists of a grouping module that aggregates similar input–output data, a classification module that derives classification conditions, and a condition extraction module that identifies pairs of partitioning targets and thresholds. This approach enables boundary conditions to be represented as general condition expressions, facilitating interpretation by LLMs and engineers.Experimental results demonstrate that the proposed method estimates boundary conditions in a more general and interpretable form than conventional methods. Furthermore, a certain level of agreement with boundary conditions designed by test engineers is achieved, indicating its practical applicability.