Goal

I have had many different ideas for new theme park attraction ideas. While I find my own ideas to be enticing, I truthfully had no information about the quality of my proposed attraction ideas and how they would actually fare if constructed and implemented.

So, I went out to try to answer some questions. Specifically, if I were to create an attraction, how popular would it be? What would be some comparable attractions I could look at to visualize how mine might turn out?

Available Information

I gathered wait time information from laughingplace.com for 8 different theme parks from December to the unfortunate closures in March due to COVID-19. The parks I looked at were the 4 in Walt Disney World, 2 in Disneyland Resort, and the 2 in Universal Orlando Resort.

Because of the unsuspected shutdown of all the parks, I have yet to gather enough data to feel very confident in my predictions, and they should improve as time comes.

I also gathered demographic information on the attraction, such as what type of attraction, whether there was a specific IP theme, the time since the attraction first opened, and the location of the attraction.

Methodology

Using the aforementioned data as well as some seasonal inputs such as day of week and whether there is a special park event or season, I used an Extreme Gradient Boosting (XGBoost) model to predict the average wait time to expect for that attraction.
XGBoost The model predicts the average wait time for a specific bucket of time of day.

Then, using some of the demographic information as well as the average wait time for the attraction, I calculated the distance between the attractions to create Attraction Similarity Scores.

Distance between variables

Distance between variables

Dashboard

https://davidiwan1.shinyapps.io/WaitTimePrediction

Create Your Attraction: For the check boxes, the user will select which of the available features they want in their new attraction. Please don’t select more than one of the same type (i.e. Park, Land, Theme)! For Height Requirement, the user can select the minimum height they would like to have on this attraction. I’d generally suggest that Thrill Rides have Height Requirements of at least 40, as in order for the attraction to pass the safety tests it will need a safe Height Requirement! For Days After Open Date, the user can select how many days after the attraction opens they want to see these wait times. How new the attraction is weighs heavily on how long the wait times will be, so the user can choose when in the future they want to check in on their creation. Feel free to be creative and keep playing around in order to see how the following change! And, most importantly, have fun and keep those creative muscles working!

Predicted Wait Times: You will see a graph of the predicted average wait time of your attraction and how that changes throughout the day.

Most Similar Attractions: Using the predicted average wait time and the inputted demographic information about the attraction, we generate the top 5 most similar attractions at the 8 theme parks to the user creation. Now you can see which attractions compare the closest to your new creation to help visualize how it may look and perform at the park!

Current Limitations

There were a few limitations to this analysis that led to a wide degree of uncertainty. I had already mentioned how I don’t have a complete amount of data due to the COVID closures. Also, I will look to adjust for any mass behavior changes once the parks reopen.

I also gathered data at discrete times as opposed to continuous marks that Disney and Universal have, and that information would allow for more accurate wait time measures and the time of day trends.

Another impact is that I don’t have enough information on how wait times are when the attraction first opens and how that smooths over time. My sample only really dealt with one real opening that has a standard queue (Mickey and Minnie’s Runaway Railway) and a few others even in proximity to their opening date, so the more information I gather in the future the more accurate my handling of those attractions and the impact of the proximity to the opening date.

Future

There are a few things I want to look into in the future.

One would be predicting the course of the year. As I get more data on the seasonal effects, I’d like to be able to predict wait times over the course of the year and see the impact of certain seasons and special events. I am unsure as to when I will get a good sample for that due to the closures and potential adjustments around COVID response.

I also want to implement uncertainty and prediction intervals if I’m legitimately stating a prediction on how I expect the attraction to perform in the park. Average wait time is an adequate representation for popularity for this specific use case but if I am stating a legitimate projection to act upon I want to include the intervals to show the range of probable outcomes and the level of uncertainty surrounding them.

There’s one other fascinating trend I’d like to try to implement. The parks are experimenting more with alternate queue options, most notably boarding groups for attractions such as Rise of the Resistance, Race Through New York, and the experience at Volcano Bay. That may become more widespread so I would love to be able to measure and predict that as well. For this specific project I omitted alternate queueing options but I do want to include those in the future as the parks may go more towards that style of queue, as well as the act that it’s relevant information on attraction popularity that I’d rather not ignore.