New technologies can improve the inclusion of deaf (and hearing loss) people in different scenarios. In the present work, a classification of the Colombian sign language alphabet was implemented. For this, the employment of the media-pipe hands pose tool was used to feature extraction process. Then, three machine learning models: support vector classifiers, artificial neural networks and random forest, were trained to determine the best proposal. Results show how a neural network with one hidden layer obtained the best performance with 99.41percent-flag-change. The support vector classifier reached an accuracy of 99.12percent-flag-change, and the worse result was achieved by the random forest model with 96.67percent-flag-change in the classification. The proposal can contribute with advances in the sign language recognition in the Colombian context, which has been worked in different approaches with more complex models to do similar classifications.