Support Vector Machines (SVM) is a machine learning tool, originally developed in the field of artificial intelligence, to perform both classification and regression. In this paper we show how SVM can be used to determine the most relevant quantities to characterize the confinement transition from low to high confinement regimes in Tokamak plasmas. A set of 27 signals is used as starting point. The signals are discarded one by one until an optimal number of relevant waveforms is reached, which is the best tradeoff between keeping a limited number of quantities and not loosing essential information. The method has been applied to a database of 749 JET discharges and an additional database of 150 JET discharges has been used to test the results obtained.