This is my next forecast for Google search interest in Michigan for 10 forms of Outdoor Recreation. These are 14 day forecasts based on a Long Short Term Memory (LSTM) neural network including variables for weather, recent search interest, time of year, day of week, and holidays. The main project page is pasted at the bottom.

So far, I’m able to provide reasonable forecasts. Some are better than others. Search interest for RVing, ATVing, and boating has generally been lower than the long-term average. As discussed during the original project, these require big-ticket purchases and, often, financing. After 2020, it was easy to finance these purchases, but recent interest is much lower. The forecasts keep pulling to the long-term average, quickly ignorning the lower recent history. This may be an area for improvment in future versions of the model.

Eventually, I’ll make improvements to this report such as comparing past and current forecasts, comparing past forecasts to actual values, and tracking the input weather variables. This will provide me with important information when I improve the actual model in the future. However, this report will never be as involved as with my Northern Michigan search interest project due to the more frequent runs.