The adoption of machine learning methods for recognizing human activities has shown remarkable results in the extraction of discriminative characteristics from sequences acquired through images. Thus, this research aims to carry out the recognition of activities carried out by students undergoing training at UniverCemig in the process of certifying their learning. This is a quantitative-exploratory research, which applied the Support Vector Machine, K-Nearest Neighbors, Gated Recurrent Unit and Long Short Term Memory techniques. As a result, K-Nearest Neighbors and Support Vector Machine achieved 91% and 96% performance, respectively, while Gated Recurrent Unit and Long Short Term Memory achieved 96% performance for both. It is believed that this research can bring an opportunity to implement the recognition of human activities in the training of students at UniverCemig, helping instructors during training and, consequently, bringing agility to the student's learning certification process.