I have just learned LSTM for one month, and I am doing a project that aims to train an LSTM model forecasting the taxi demand at "t+1" according to the taxi demand at "t", "t-1"... In particular, I am focusing on the event venue. I want to incorporate the data that could show what is happening during the event (like the score of a basketball game, or weather condition).
Now, I have about 70 taxi pick-ups time series (different days of the match), I can also have the score of each match and weather condition, and build them as time series.
I want to train an LSTM by using multiple pick-ups time series and those corresponding weather and score time series. How can I achieve this?
One problem might be that I mainly want to forecast the pick-ups after the event when the score time series already ends. As a result, the length of the pick-ups time series and score time series are different. How can I cope with this problem?
Is there any other machine learning methods that could handle this problem more effectively?
Thank you so much!