Figure 2
Feature importance for model interpretability. The present figure represents the relevance of the information used by the model to perform classification, namely the feature importance, both using avalanche transition matrix (ATM) and imaginary coherence (ImCoh). Panel (A) is a histogram of the probability of the feature importance value in ATM and ImCoh. The histogram shows a narrow distribution for ImCoh and a broader distribution for ATM, suggesting that in ATM certain edges drive the majority of information necessary for differentiating the two groups. Panel (B) is the edge representation in a chord plot, showing the importance of each edge in the classification. Finally, panel (C) is the mean importance value of each edge of a specific brain region. This representation highlights which regions mainly impact the classification.
