I$\text{'}$m a beginner in deep learning and time series analysis. I have a transportation dataset that tracks passenger flow across multiple routes$/$lines$.$ My challenge is to forecast multiple related outputs:

The data includes:

Multiple target outputs $(4$ different lines and their total$)$

Time$-$based features

Environmental factors

Economic indicators

Demographic information

Key questions:

How should I approach this multiple$-$output time series forecasting problem?

Should I build separate models for each line? Or should I create one model with multiple outputs? How to handle the relationship between individual lines and their total?

For implementing LSTM:

How to structure data when we have multiple related outputs? What would be an appropriate architecture? How to handle the dependencies between different outputs?

For evaluation:

How to measure accuracy across multiple outputs? How to ensure predictions for individual lines align with total predictions?

Could you guide me through this process with explanations and sample code, considering I$\text{'}$m new to both time series analysis and deep learning?