Feature-Based Facial Emotion State Recognition Using a Hybrid Method of Machine Learning and Deep Learning

Facial Emotion State Recognition (FER) plays a crucial role in human–computer interaction, educational technology, and intelligent service systems. Unlike conventional approaches that depend solely on either handcrafted descriptors or deep learning, in this paper, we propose a robust hybrid FER model that combines both paradigms to exploit their complementary representational strengths. Experimental evaluation on the FER2013 dataset demonstrates that the hybrid model achieves an accuracy of 73.12% and a Macro-F1 score of 71.3%, substantially outperforming both traditional machine learning and baseline deep learning approaches. Cross-dataset validation further confirms the robustness and generalization capability of the proposed method. Additionally, the model has been optimized for web-based deployment, achieving an average inference time below 130 millisecond per frame, which is sufficient for near-real-time performance in interactive applications.