Potholes are among the most common and hazardous types of pavement defects, significantly impacting road safety, vehicle durability, and infrastructure maintenance costs. This study proposes an automated multi-class pothole detection and classification approach based on the YOLOv8 (You Only Look Once version 8) deep learning model. The dataset comprises 618 real-world images collected from a public source, where each pothole is annotated into eight distinct classes based on three attributes: depth (shallow/deep), surface condition (wet/dry), and size (small/large). The YOLOv8n model was trained on 640×640 pixel images for 100 epochs, utilizing transfer learning from COCO (Common Objects in Context) weights and advanced data augmentation techniques to enhance generalization. Experimental results on the test set achieved mAP50 = 0.397, mAP50-95 = 0.226, Precision = 0.430, and Recall = 0.417, with an average inference time of 12 ms per image, demonstrating the feasibility of real-time applications. The findings confirm that YOLOv8 is effective in detecting and classifying potholes under various environmental conditions. The proposed system can be integrated into autonomous vehicles, drones, or innovative infrastructure monitoring platforms, contributing to automated road inspection and enhanced traffic safety in smart transportation systems.