Figure 3

Classification performance and extrapolation. (A) Comparison of 360 classification models (90 models per classifier and panel) regarding three performance metrics: F2 score, Recall, and ROC AUC (each dot represents a model). QDA model (1) shows the highest F2 score, QDA model (2) achieves the highest Recall, and Random Forest model (3) the best ROC AUC score. (B) Confusion matrices of the three best performing models (1) to (3) according to the performance metrics (top panels) and a random draw model (4) (lowest panel). Inset: performance comparison between machine learning models (red) and 20,000 random draws (blue) according to F2 score, Recall, and ROC AUC. The machine learning models show a significantly better prediction performance than the random draw models. (C) Predicted share of GreenLeader accommodations in Europe (NUTS-2) according to the QDA model (1). The model predicts urban centres and several regions in West and North Europe to have the high shares of sustainable tourist accommodations