Fig. 5. Behavioural effects of β-event neurofeedback training.
a Classification accuracy of the SVM models (n = 3 repetitions) trained on epochs with correct labels vs. models trained on epochs with shuffled labels to infer chance levels. SVM models achieved above chance accuracy (18% increase, 0.6 vs. 0.51, two-tailed Welch t-test, t(4) = 3.284 p = 0.03). b Temporal attention extracted from the SVM models on correctly predicted true samples. The time course of the mean normalized attention over all true positive epochs ± SEM indicates that movements correlated with beta bursts occurred primarily shortly before burst initiation (linear regression, ρ = 0.87, p = 2.8 × 10−23). c Examples of the model attention in individual epochs. The temporal (left panel) and spatial (right panel, including a heat map overlay for the video frame with highest attention, marked with a red x in the temporal profile) attentions of the model follow similar patterns for representative epochs of each rat. The attention implies that the movements of frontal body parts, shortly before burst initiation, were most informative to predict the epoch class.