This study proposes a novel deep learning model integrating Convolutional Neural Networks (CNNs) with Squeeze and Excitation Blocks (SEBlock) and Optical Flow for improved human action recognition. The SEBlock enhances the network’s ability to recalibrate channel-wise feature responses, allowing the model to focus on more discriminative motion patterns. At the same time, Optical Flow captures temporal dependencies across frames, ensuring robust motion representation. To evaluate the effectiveness of our approach, we utilize the KTH dataset, which consists of six human actions (walking, jogging, running, boxing, hand waving, and hand clapping) performed by 25 subjects under varying conditions, including different lighting scenarios and viewpoints. This dataset provides a challenging benchmark for assessing action recognition models due to its diversity in execution styles. Our proposed model achieves a classification accuracy of 92.13%, outperforming conventional 3D-CNN (90.2%) and LF+SVM (71.72%) methods. Notably, it attains 100% accuracy in Boxing and Walking, demonstrating superior capability in recognizing both dynamic and repetitive actions. The key contributions of this research include the development of an enhanced CNN-based model that effectively captures spatiotemporal dependencies, achieving state-of-the-art performance on the KTH dataset, and providing insights into the integration of SEBlock and Optical Flow for improved action recognition.