I am trying to build a model that identifies anomalies based on 50+ variables. The data however is generated in streaming mode ( with a fixed time interval of 5 seconds). How do I impute missing values (for all variables at a given time step) in such multivariate streaming data? Also, is it reasonable to assess and check for anomalies for those timesteps which are generated through 'imputation' and do not represent real values?