Deep learning-based restaurant scoring system based on customer's facial expression

In today's automated dining environments, collecting authentic customer feedback has become increasingly challenging due to reduced human interaction. Traditional feedback mechanisms, such as surveys and touchscreen kiosks, are often underutilized and fail to capture genuine emotional responses. This project introduces a non-intrusive, real-time facial expression recognition system designed to assess customer satisfaction based on facial cues. Utilizing pre-trained deep learning models such as EfficientNet and MobileNet, the system classifies customer expressions into three categories: satisfied, neutral, and unhappy. A camera captures facial data post-meal, and the model processes it instantly to provide feedback without requiring active user participation. The system is optimized for kiosks and embedded devices, offering a contactless, hygienic, and intelligent alternative to conventional methods. Experimental results demonstrate that EfficientNet achieves a classification accuracy of 96.2%, outperforming MobileNet's 93.7%, while maintaining real-time processing capabilities suitable for deployment in commercial restaurant settings.