Figure 2.

Machine Learning Analysis on Cross-Disciplinary Data. Natural Language Processing (NLP) data were aggregated with demographic, neuroimaging, and psychological data to predict Toronto Structured Interview for Alexithymia (TSIA) scores. Candidate features (n = 92) were extracted using univariate filtering with Spearman correlation method, Principal Components Regression (PCR) model with three components was selected as it had the best performance over evaluation metrics assessed in Leave-One-Out Cross-Validation (LOOCV) (R² = 0.86; Root Mean Square Error—RMSE = 0.35) (A,B). Variable importance is reported (C1,C2). Namely, 85 candidate features resulted to belong to NLP, 3 to psychological measures, 2 to neuroimaging data, and 2 to demographic data. The variable importance of neuroimaging data showed that the increased left temporal pole thickness (orange colored area, D1) predicted increased TSIA scores, while increased right isthmus cingulate thickness (orange colored area, D2) predicted decreased TSIA scores. In (D1,D2) left is left.