Statistical learning models are proposed for the prediction of the probability of a spike in the electricity DART (day-ahead minus real-time price) spread. Assessing the likelihood of DART spikes is of paramount importance for virtual bidders, among others. The model's performance is evaluated...