Face Presentation Attack Detection (FacePAD) is a crucial task for improving the security of facial recognition systems and preventing them from potential spoofing attacks. FacePAD methods come with a few challenges, such as being unable to generalize, inefficient for computation, and inability to tackle most complex attacks like video replays and 3D mask spoofing. These constraints limit the performance of existing methods, primarily in practical applications that demand both precision and efficiency. To overcome these obstacles, this paper presents an improved FacePAD method which consists of spatio-temporal feature extraction and a lightweight Convolutional Neural Network (CNN) framework, MobileNetV3. With better resistance against presentation attacks, this approach extracts static and dynamic facial characteristics through real time video sequence. Experimental results clearly show that the proposed method systematically outperforms their approach in terms of accuracy of classification, and shows better performance over a variety of spoofing attack types. The method is validated on some established datasets, such as Replay-Attack and Replay- Mobile and ROSE-Youtu, and it consistently outperforms other state-of-the-art methods for detecting sophisticated presentation attacks. What this is leading to is potential improvement for biometric secure systems.