Full Model Selection in Huge Datasets and for Proxy Models Construction

Abstract

Full Model Selection is a technique for improving the accuracy of machine learning algorithms through the search of the most adequate combination on each dataset of feature selection, data preparation, a machine learning algorithm and its hyper-parameters tuning. With the increasingly larger quantities of information generated in the world, the emergence of the paradigm known as Big Data has made possible the analysis of gigantic datasets in order to obtain useful information for science and business. Though Full Model Selection is a powerful tool, it has been poorly explored in the Big Data context, due to the vast search space and the elevated number of fitness evaluations of candidate models. In order to overcome this obstacle, we propose the use of proxy models in order to reduce the number of expensive fitness functions evaluations and also the use of the Full Model Selection paradigm in the construction of such proxy models.