Forecasting error (via RMSE) of various models and features using the Twitter dataset.

The models used include VARMA and SVR, whereas the features used include the raw social dynamics (Baseline), level-1 activation vectors (H1), level-2 activation vectors (H2), and both levels of activation vectors (H1 + H2) of a 2-level RCBM. The forecasting accuracy using a 1-level RCBM with a doubled number of filters is denoted as (H12).