This is my ninth installment for forecasting search interest for ten forms of outdoor recreation in Michigan. This installment provides a final cross validation comparison between two of the three models I developed during previous installments:

1. Hankel/VAR: This more traditional approach from Installment 6 employs time series decomposition and statistical models.
2. Long Short Term Memory (LSTM) Model: This approach from Installment 8 is a more advanced approach that employs a neural network.

Unsurprisngly, the best model depends on which of the ten forms of outdoor recreation you are modeling. It’s important to note, however, that there are inumerable possible iterations of these models which means that different approaches or evaluation over different time periods could easily change outcomes.

My next installment will discuss sourcing weather forecasts from accuweather.