The benefits of machine learning are undeniable in most, if not all the aspects of human activity. From weather forecasting to classifying a tumor as benign or malignant the use of machine learning speeds up and facilitates solving the problem at hand. However, the deployment of the most suitable machine learning model for the problem at hand is a time consuming process that requires knowledge gained through experience and continuous practice on the subject. The aforementioned obstacles can be removed by means of automated machine learning which, as the name suggests, attempts to automate the development process of machine learning models, and thus the benefits of the former to become broadly available. Contributing towards that end is the field of meta-learning, which studies the performance of different machine learning models in a wide range of tasks and uses this experience to "predict" the most suitable model for a given task, avoiding the process of trial and error. This thesis deals with the task of automating machine learning by employing meta-learning techniques specifically for regression problems. The aim of this work is the development of a system able to decide the exact optimal values for the hyperparameters of three algorithms, given the data to process. In that way a great deal of time is saved during the deployment of the model and the use of machine learning becomes available to more groups of people.