I have the following task - predicting the next 12 hours of PM10 particles based on historical data of previous 24 hours of PM10, O3 (ozone), CO (carbon monoxide), and others (not included) using RNN's.
When I plot the present vs missing values in the dataset, this is what I get (blue = present, white = missing):
My question is, should I impute the missing values in the dataset?
If I do end up imputing the values, won't that have an effect on the performance of the model, since PM10 in my case is both input and target variable?
On the other hand in case I drop the values, then I will have to make sure that the previous 24 hours and the following 12 hours for each timestamp are consecutive timesteps, correct? (there should be no overlap because of truncation).