Project Portfolio Management is relevant for the growth of companies since it favors planning. Project Portfolio Management manages the resources to plan, control, and execute projects and obtain the strategic objectives of the organizations. In Project Portfolio Management, a large amount of data is forged, important for planning new projects in companies; therefore, the need arises to create models that help process and interpret the data. In this context, Machine Learning is presented as a technological enabler that allows a system, by itself and in an automated way, to learn to discover trends, patterns, and relationships between data; it is an engine of digital transformation of business and that organizations are embracing. Therefore, this article aims to compile and review proposals made to implement machine learning in the management of the project portfolio and apply algorithms that allow the development of models that help in the management and evaluation of projects to be developed in a Software Factory. The CRISP-DM methodology is applied to process the data of costs, times, and types of Projects; the Python programming language is used, the dataset corresponds to a Software Factory. The results validate the models implemented using Machine Learning algorithms, such as regression and decision trees, and thereby obtain the best model for predictions, establishing the correlation between variables and the benefit to be achieved. It is concluded, the implementation of Machine Learning improves the IT Project Portfolio Management, helping to identify which projects are more profitable and beneficial.