A Practical Cost-Benefit Analysis of Pre-trained Model Selection

This paper provides a systematic framework for selecting pre-trained computer vision models in enterprise applications by evaluating YOLO, MobileNet, ResNet, and EfficientNet across accuracy, speed, and cost dimensions. Through comparative analysis, we demonstrate cost variations up to 300% between models with similar accuracy, and provide actionable recommendations for cloud versus edge deployment strategies. A retail case study shows 90%+ cost savings through optimized model selection.