In Mexico, the increasing number of vehicles on the road has led to a rise in accidents, often influenced by driving errors, distractions, and external factors. This problem is especially evident among transport drivers, who face long working hours without adequate breaks, making them more vulnerable to incidents. This research explores models based on convolutional neural networks (CNN) to monitor and understand drivers’ behaviors and identify ten everyday driving activities. The models are trained on data that includes images of driving activities, which have been pre-processed to facilitate analysis. The objective is to identify and classify proper and distracting driving activities, looking for the best deep learning technique. The research is based on the hypothesis that Convolutional Neural Networks are suitable for identifying dangerous driving activities. The StateFarm Distracted Driver Detection database is used for model training and validation. However, a new database of driving activities is recreated using a methodology that involves image acquisition, data preprocessing, model training, execution, validation, and testing in real-world environments in Mexico. The results were evaluated using accuracy, precision, and confusion matrix metrics, achieving 96.80% accuracy in the tests performed.