Daily Austin Airport Wait Time Predictions
I made a model to predict how busy the airport is going to be.
During my Data Science Masters, I took a class called Advanced Predictive Models. I am applying one method I learned from that class to the Austin Airport TSA passenger counts.
Time Series Prediction with SARIMA Modeling
For these daily predictions, I use a form of time series modeling called SARIMA. SARIMA is an acronym for:
A SARIMA model takes in parameters for each of these sub-models and combines them to make predictions. It is typically written as:
SARIMA =(p, d, q) x (P, D, Q, S)
S is the period of the seasonal parameter, in this case we’re using 7 to account for the day-of-week effects of passenger volumes.
To pick parameters for SARIMA, we simply do a grid-search of P,D,Q,p,d,q parameters to see which has the best performance in a statistic called AIC.
The most optimal found for grid search in space 0 to 2:
SARIMA =(2, 1, 2) x (0, 1, 2, 7)
What this is good for
This model is great at including the seasonal and day-of-week effects on passenger volumes. It is not going to do a great job seeing one-off holidays like Memorial Day Weekend or big events like ACL fest.
The data behind these predictions gets updated on a roughly two week period. This can cause some lag, especially big changes that may have recently happened.
I’m using Google Cloud Functions to train my model, do my forecast, create the image, and tweet the results. I used this guide.