Mining Auto-generated Test Inputs for Test Oracle

Abstract

A Search-based test input generator produces a high volume of auto-generated test inputs. However, manually checking a test oracle for these test inputs is impractical due to the lacking of a systematic way to produce corresponding expected results automatically. This paper presents a mining approach to build decision tree models containing the estimated expected results for checking a test oracle. We first choose a subset of the auto-generated test inputs as a training set. Then, we mine the training set to generate a decision tree from which the estimated expected results can be retrieved. For evaluation purpose, we have applied our approach to two legacy examples, Triangle and Next Date. Our controlled experiments have shown that the mining approach is able to generate highly accurate behavioral models and achieve strong fault detectability.