Human Activity Recognition (HAR) using wearable devices has gained increasing attention due to its potential for real-time, non-intrusive monitoring. This work presents a HAR system specifically designed for educational environments, leveraging Inertial Measurement Unit data from smartwatches. We introduce a custom dataset comprising classroom-related activities indicative of student engagement, and develop a neural network-based model capable of classifying these activities in real time. Experimental results demonstrate approximately 80% accuracy in recognizing continuous gestures such as typing, writing, and drawing, and around 72% accuracy for instantaneous gestures like raising a hand or drinking. These findings suggest the feasibility of using smartwatch-based HAR systems for enhancing personalized learning and classroom management.