<p>Facial Expression Recognition (FER) has garnered significant attention due to its applications in human-computer interaction, security, and healthcare. This paper presents an enhanced FER method by integrating the Coordinate Attention mechanism with a lightweight Convolutional Neural Network (CNN), specifically MobileNetV3. The Coordinate Attention mechanism from Hou et al. injects accurate spatial data into attention processing to help the model better focus on important facial characteristics. The model adopts MobileNetV3 since this architecture successfully balances between performance speed and computational power requirements. The proposed model receives evaluation through testing on three data sets JAFFE and CK + and FER2013. The tested FER techniques demonstrate increased performance measurements compared to current standard FER processing methods. The combination of Coordinate Attention with MobileNetV3 produces a model with 99% success rate. On JAFFE the model reaches 99.28% accuracy and processes CK + with 99.10% precision and reaches 97.22% accuracy on FER2013. The proposed method demonstrates successful and precise operation across different real-world datasets through its achieved results. This paper introduces the combination of Coordinate Attention and MobileNetV3 as the primary contribution to address FER. The evaluation of the proposed method on various benchmark datasets. The exploration of the effect of attention mechanisms on lightweight CNN models. “These findings provide valuable insights that can help improve the efficiency and precision of FER systems in real-world applications requiring facial expression analysis.</p>