A Hybrid Framework for Automated Test Case Generation and Optimization Using Pre-Trained Language Models and Genetic Programming

Abstract

Automated test case generation plays a pivotal role for software quality, reliability, and efficiency, while traditional methods such as rule-based and random test generation often produce invalid or redundant test cases. With search-based algorithms like Genetic Algorithms (GA) and Particle Swarm Optimization (PSO), however, both approaches run into significant obstructions due to the complexity of software structures, which limits their effectiveness. To solve these problems, this work demonstrates a Hybrid Framework for Automated Test Case Generation and Optimization, combining Pre-Trained Language Models (PLMs) with Genetic Programming (PG). The PLMs generate an assortment of context-aware test cases by analysing software requirements, whereas the GP optimizes test cases in terms of fault detection, coverage, and execution efficiency. The experimental results that exemplify the PLM-GP method show it performing much better than traditional and PLM-only methods, with 92.5% statement coverage, 85.3% branch coverage, and 89.1% rate defect detection, while decreasing redundancy to 12.5%. Along with that, by comparing the performance with PLM-EA, PLM-GP shows further increase out of the achieved ones: higher accuracy of 0.98, higher diversity of 0.91, and higher 0.96 in coverage with an increased fitness score of 0.97. Intelligent test case generation combined with the pre-trained language models' genetic programming optimizes the effectiveness and efficiency of software testing. The study shows that PLM-GP is a powerful and scalable methodology for automated test case generation, improving fault detection, test coverage, and overall quality above the current techniques. Combining or reducing redundancies, as well as optimizing test execution, enhance software reliability while reducing human effort during the testing process.