Executive functions (EF) are a class of processes critical for purposeful goal-directed behavior. Cognitive training (i.e. the adequate stimulation of EF) has been studied and applied for the last 25 years in hugely diverse backgrounds. In spite of the accumulated evidence of its positive impact in cognition, there are still reports in the literature that claim that the potential benefits of training are not generalizable. Recently, research considers individual differences as one of the possible causes of these inconsistencies. Is it possible to build one training protocol that benefits everyone? Or is it time to stop considering individual differences as inevitable experimental noise and, instead, use them as information that can guide us towards finding the best possible strategy for each person?
In this study we use Machine Learning algorithms to identify and describe possible subgroups of individuals that will (or will not) benefit from a certain stimulation. The algorithms were built using data from a cognitive training intervention (N=73 6 y.o.) run with a set of computerized games aimed at training and measuring EF (www.matemarote.org.ar).
We present a Nearest Neighbors classifier that successfully predicts whether a subject will benefit or not from a fixed training approach based on his/her performance in previous cognitive tests. In the long term these algorithms will allow us to individualize training protocols in order to maximize the stimulation for each child.