[Discovery Day Q1 2023] Deep Learning for Parking Revenue Prediction

[Charlie-Henry]

Repo. I managed to get my data ready beforehand this time, wow!

The challenge: It can be hard as a department to predict revenue from parking besides typical methods (like extrapolating how much was generated in a previous month in a previous year to a current year) that can be significantly inaccurate. It'd be interesting to see what the trends are in parking demand and maybe see how we could generalize a model that would predict demand at new parking spaces (like South Congress?).

The idea: Deep learning can be used in time series forecasting, usually in the context of predicting stock markets (a completely unpredictable space usually). I might follow this DeepAR approach claiming:

DeepAR outperforms the standard ARIMA and ETS methods 👀

[Charlie-Henry]

I got a really basic 2-layer linear model and got some half decent results. I'd like to dig deeper on this later and try some different model architectures.