This study investigates the influence of dataset size and the number of training epochs on the performance of YOLOv8 object detection models. Three model variants: YOLOv8s, YOLOv8m, and YOLOv8xl, were trained using five datasets of increasing size (200 to 1000 images) and five epoch configurations ranging from 10 to 50 epochs. Model performance was evaluated using standard validation metrics, including mean Average Precision at an IoU threshold of 0.5-0.95 (mAP50-95), precision, and recall. The results indicate that increasing dataset size and training duration generally improves detection performance, with performance gains diminishing beyond approximately 40 training epochs. Model generalization capability was further assessed using a separate test set consisting of 32 manually annotated images, employing a similarity coefficient that integrates localization accuracy, classification correctness, and detection completeness. Additional analyses of true positives (TP), false positives (FP), and false negatives (FN) were conducted at confidence thresholds of 0.25, 0.50, and 0.85 to examine detection robustness under varying confidence constraints. Among the evaluated models, YOLOv8m provided the best balance between detection accuracy and computational efficiency. YOLOv8s, although less accurate, remained well suited for real-time inference and deployment on resource-constrained systems. In contrast, YOLOv8x did not exhibit consistent performance improvements relative to its increased computational complexity.