Recognizing and responding to students’ emotional states during classes is crucial for optimizing learning outcomes, yet it remains challenging for educators. This study presents the development and implementation of a real-time emotion recognition system using convolutional neural networks (CNNs) to analyze students’ facial expressions during in-person classes. Trained on the FER2013 dataset, the system classifies seven distinct emotions with 85% accuracy. An experiment with 20 university students aged 18–25 compared emotional responses across three teaching methodologies: collective, group, and experiential. Experiential teaching elicited the most positive emotions, with 50% of expressions classified as happiness, while collective teaching generated more negative responses. Statistical analysis revealed a significant positive correlation between happiness and academic performance (r = 0.65, p < 0.01) and a negative correlation between fear and performance (r = −0.54, p < 0.05). The system provides educators with quantitative emotional data, enabling real-time adaptation of teaching strategies and retrospective analysis of class dynamics. This research contributes to AI-enhanced education, offering insights into creating more responsive and student-centered learning environments while addressing privacy and data protection considerations throughout the study.