Such differences can be exploited in order to make predictions on the strategy that a player is going to adopt, based on the on-line analysis of hyper-brain networks constructed from data recorded in the decision-making process. For each frequency band, we have implemented a non-linear classifier, more specifically a Multi-Layer Perceptron (see Methods S2 and Figure S5 for details), using 21 couples for training (6 networks per couple, each graph corresponding to one of the 6 different strategies, for a total number of 126 networks), and the remaining 5 couples (30 networks in total) for validation. The classification is based on the values of the Z-scores of efficiency, divisibility and modularity, and not on the actual values of the measures themselves. In fact, for each couple, the Z-score of a graph measure provides its deviation from the average value computed over all hyper-brains of the same couple. The accuracies obtained by the classifiers during validation process, i.e. the number of hyper-brain networks classified correctly as DD or non-DD out of the 30 validation patterns, are respectively: 27, 22, 26, 24 for the Theta, Alpha, Beta and Gamma frequency band.
