Comparative Analysis of Deep Learning Models for Validating Use Case Diagrams

In the ever-evolving world of technology, the validation of use case diagrams is essential to ensure the reliability and consistency of software applications. An effective validation approach plays an important role in optimizing system quality. This study adopts a comparative approach to analyze the performance of three major models in object detection, namely Detectron2, YOLOv5, and YOLOv9, to validate use case diagrams. These three models were trained using an open repository dataset, Roboflow "Use Case Diagram Checker Computer Vision Project". Through a series of careful experiments, we evaluated and compared the three models based on relevant performance metrics, such as precision, recall, and AP50. Our analysis results show that YOLOv9 outperforms the other models with significant improvements in detecting objects in user case diagrams. Although YOLOv9 shows superior performance, we also consider other aspects such as model speed and complexity. This research not only provides deep insight into the relative performance of each model in the context of user case diagram validation but also provides valuable insights for practitioners and researchers in choosing the approach that best suits their needs. Thus, the contribution of this research is highly relevant in the development of reliable and efficient software systems, as well as being a practical guide for the selection and implementation of object detection technologies in the context of user case diagram validation.