A Metaheuristic and Neural Network-Based Framework for Automated Software Test Oracles Under Limited Test Data Conditions

Abstract

With the growing complexity of modern software systems, the demand for effective and efficient testing techniques has become an important aspect of the software development process. Software Test Oracles (STOs) play a vital role in testing by determining whether a program behaves as expected for a given input. This study introduces a novel automated STO framework that utilizes metaheuristic algorithms and ML techniques to enhance testing precision and reduce the testing cost. The proposed approach begins with generating coverage-based test data using a hybrid of the Imperialist Competitive Algorithm (ICA) and Genetic Algorithm (GA). The initial test data is optimized using Hamming distance to address redundant test data and improve efficiency. This reduced dataset is used to train a multi-layer perceptron and to create an STO that accurately predicts the software under test’s expected output. The oracle was validated using both original and mutant versions of standard benchmark programs. Additionally, an automated platform has been developed to support Oracle creation, test case generation, and validation. Experimental results demonstrate that the proposed STO attains high accuracy (96.70%) and recall (98.63%), highlighting its effectiveness when a limited quantity of test data is available.