The level of student engagement is crucial in determining educational success and significantly influences learning outcomes. Traditional methods for detecting engagement are prone to subjectivity and limited reliability. Unimodal systems that only detect facial cues often miss scenarios where students are cognitively engaged but not facing the lesson. To address these limitations, we propose a two-level verification approach that combines head pose estimation with facial expression analysis. The first level uses a Convolutional Neural Network (CNN) trained on Head Pose Image Database to classify head poses into four categories, achieving an accuracy of 89.36%, indicating visual attention. The second level utilizes pre-trained models, ResNet50 and EfficientNet, to analyze facial expressions using the DAISEE (Dataset for Affective States in E-Environments) dataset. Among these, ResNet50 shows the highest performance with an accuracy of 62.27% in categorizing engagement states. This two-pronged approach offers a more robust and accurate method to assess student engagement.