<p>Facial emotion recognition (FER) plays a pivotal role in non-verbal communication, particularly in enhancing interactions between humans and machines. While recent advances in machine learning and deep learning have significantly improved the accuracy of FER systems, many existing models remain computationally intensive and unsuitable for deployment on mobile or embedded devices. In this study, we present a lightweight and efficient FER framework based on MobileNetV2, fine-tuned to address challenges such as class imbalance and low-resolution grayscale input. The MobileNetV2 architecture, originally pre-trained on the ImageNet dataset, was extended with custom layers to enhance performance and was evaluated on two widely used benchmark datasets—FER-2013 and Cohn-Kanade+ (CK+). Our model’s effectiveness was assessed using comprehensive performance metrics, including accuracy, precision, recall, F1-score, Area Under the Curve, and Root Mean Square Error. The fine-tuned MobileNetV2 achieved an accuracy of 92.39% on the CK+ dataset and 66.68% on the FER-2013 dataset, demonstrating a favorable balance between recognition performance and computational efficiency, making it suitable for real-time and resource-constrained applications.</p>