Metamorphic Testing Framework for Deep Learning-Based Smart Traffic Signal Management

Abstract

Deep learning models are central to modern in- telligent traffic systems but remain difficult to validate due to the absence of explicit test oracles. This study presents a Metamorphic Exploration (ME) framework to systematically evaluate the robustness and reliability of deep learning models in real-world traffic scenarios. The approach is demonstrated using the YOLO (You Only Look Once) model within a Smart Traffic Signal Management System. Metamorphic Relations (MRs) are defined based on key testing properties—robustness and effec- tiveness—using metrics such as accuracy, precision, recall, and class loss. Experimental results on both public and real-world datasets show that ME effectively measures model reliability and performance consistency under diverse traffic conditions.