Personal protective equipment (PPE) usage on construction sites is a significant factor in minimizing accidents and enhancing the standard of occupational safety. This paper offers a comparative study of various YOLO models for object detection of PPE on construction sites. The dataset includes 2,224 images of construction sites sourced from Kaggle and labeled manually using Roboflow. Five classes of safety equipment were considered: boot, helmet, gloves, vest, and goggle. After labeling, the dataset included 2,305 boots, 1,911 gloves, 712 goggles, 7,647 helmets, and 5,521 vests. The dataset was split into training, validation, and testing sets. Four object detection models-YOLOv5, YOLOv8, YOLOv10, and YOLOv11-were trained and tested using precision, recall, and mean Average Precision at IoU 0.5(mAP)0.5) as evaluation criteria. Among the tested models, YOLOv5 performed best with precision of 0.7116, recall of 0.6361, and mAP0.5 of 68.28%. The findings indicate that YOLOv5 has a better balance detection ability for safety equipment compared to the latest YOLO models in this experimental scenario. This study offers a significant contribution to the automation of construction site monitoring in determining the best YOLO model for PPE detection based on the conditions of the given dataset.