Attention Deficit Hyperactivity Disorder (ADHD) and Autism Spectrum Disorder (ASD) hinder learning and productivity among students in class. Traditional monitoring methods, such as teacher observation, lack objectivity, necessitating AI-enabled solutions. This study discusses the performance of facial emotion recognition (FER) models, that is, Xception, in identifying emotional responses from children with ADHD and ASD. The study used an ASD dataset (833 images) and an ADHD facial emotion dataset (6,000 images). Pre-processing was done by resizing, normalization, and augmentation. The modified Xception model achieved 68% test accuracy on ASD dataset and 57.34% test accuracy on ADHD test dataset. Results indicate its prowess in processing joy and sadness but still challenges with fear and anger recognition. Confusion matrices and classification reports indicate class imbalances affecting model generalization. Ethical issues such as privacy and data protection are still central to real-world deployment. The study highlights AI potential to assist with neuro-divergent students yet calling for properly balanced datasets along with more honed model training.