Figure 4
Model performance as a function of the number of available training points. For the six analyzed countries and four different forecasting horizons (1–4 weeks), the figure shows the averaged differences between the MSE of the naive approach and the MSE of the forecasting model across the different splits, as a function of the size of the training set. Error bars correspond to the relative standard deviation. The green area indicates where our model outperforms the naive one, the red area indicates the opposite.
