Pembelajaran Mesin Untuk Deteksi Helm Keselamatan Menggunakan Algoritma YOLOv8

Implementation of Occupational Safety and Health (K3) in construction projects is an effort to create a safe and healthy work environment, free from work accidents and work-related diseases. Based on the International Labor Organization (ILO) report in 2019, more than 395 million workers worldwide experienced non-fatal work injuries. In Indonesia, BPJS Employment reported 370,747 work accident cases in 2023, with an increase of 19.7% from the previous year. As many as 32.12% of these incidents were caused by not using personal protective equipment (APD), including safety helmets. This research aims to build a model through machine learning to automatically detect the use of safety helmets using the YOLOv8 algorithm. The YOLO algorithm is known to have good detection speed and is suitable for complex construction environments. The model evaluation results show that Average Precision reached 78% for all classes, 82% for the APD class, 72% for the non-APD class, and 80% for the non-safety class. The resulting precision value was 77.7% and the recall value was 60.7%, with a mAP (mean Average Precision) level reaching 68.9%. The model training time lasted 1 hour 7 minutes.