Task 1: Add activity patterns to synthetic population

[HSalat]

• [x] Task 1a: Explore current data
  • [x] TUS (public) (MVP)
  • [ ] NTS (private, available but needs understanding for Tech Plat use) (for future)
• [ ] Task 1b: Add activity chains to population (scoping)
  • [x] Step 1: Statisical matching (Propensity score matching) OR categorical matching (MVP) - AS DONE IN SPC
  • [x] Step 2: Convert matched dataset to activity chain
  • [ ] Future improvements
  • [ ] Markov chain model (for future)
  • [ ] Bayesian network model (for future)
  • [ ] Task 1c: Validate activity sequences
  • [ ] sequence alignment methods (for future)

-- Comments (HS):

• 1a: TUS appears ill-fitted compared to NTS for the task; -> NTS-SPC merge
• 1b/1: I'm skeptical of PSM from previous merge attempts. Identify (1) matching variables and (2) method
• NTS sample (non special access): Trips contain low numbers of NAs and for non-important variables only (~1.2% of start times missing). It seems only almost complete cases were kept in the sample -> how do we know how exhaustive the data is for one particular individual?

[Hussein-Mahfouz]

1b/1: I'm skeptical of PSM from previous merge attempts. Identify (1) matching variables and (2) method

In terms of methods, I need to:

• Understand PSM
• Compare PSM to the statistical matching approach mentioned in Synthetic population and travel demand for Paris and Île-de-France based on open and publicly available data and choose one

Optional / Future Enhancements:

• Bayesian Networks: paper, code