Lightweight Deep Learning for Efficient Test Case Prioritization in Software Testing Using MobileNet & TinyBERT

Abstract

Software testing is crucial for ensuring the
reliability and robustness of complex systems, with
test case prioritization (TCP) playing a vital role in
optimizing testing efficiency. Traditional methods,
such as heuristic-based and genetic algorithm-based
approaches, often suffer from high computational
costs and suboptimal prioritization, limiting
scalability in large-scale software projects. To
address these challenges, this work proposes a
lightweight MobileNet + TinyBERT framework for
efficient test case prioritization. Unlike traditional
models, this approach integrates TinyBERT
embeddings for textual test case descriptions and
MobileNetV3 for Control Flow Graph (CFG)
analysis, enabling better prioritization with minimal
computational overhead. The proposed method
achieves higher test case coverage (95%), improved
execution efficiency (92%), and better defect
detection accuracy (97%) while reducing
computational overhead (55%) compared to
Advanced Genetic Algorithms (AGA). These
improvements result in a more scalable, faster, and
accurate TCP process. The comparison with AGA
highlights superior performance in test case ranking,
ensuring enhanced software validation. By reducing
computational costs and increasing accuracy, this
framework significantly improves the efficiency and
reliability of software testing. This work contributes
to advancing deep learning-based software testing,
demonstrating the potential of lightweight AI
models in large-scale TCP applications. Future
research will explore self-supervised learning
techniques to further enhance defect prediction and
test prioritization in dynamic environments.