CNNs don’t have the assumption that history is complete: Unlike RNNs, CNNs learn patterns within the time window. If you have missing data, CNNs should be useful . In a way, CNNs can look forward: RNN models only learn from data before the timestep it needs to predict.