Magic Kingdom Park
Orlando, USA
park.fan · forecasting model
Fancast
It reads millions of live wait times to predict how busy a park will be up to 365 days ahead — and grades itself, in the open.
- ±8.8
- min avg error
- 134+
- parks
- 365
- days ahead
- Daily
- retrained
Fancast is our in-house forecasting model — the part of park.fan that looks into the future. The name? Shameless but systematic: fan as in park.fan, cast as in forecast. A weather report for queues, basically.
And because we only trust numbers that have to prove themselves, Fancast does something most models quietly avoid: it grades itself. Every prediction is later checked against the wait time that actually happened — in the open, on this page. Cheating pointless.
In short: Fancast is not a fortune teller with a crystal ball. It is a stubborn statistician that gets tutoring every night and has to re-sit the exam every morning. A weather frog that fact-checks its own weather.
01
The report card
How good is Fancast really?
Enough preamble — here is the grade, live and unvarnished. Fancast pulls these numbers from its own dashboard right now; they shift the moment the model retrains tonight.
AI-Powered Predictions — How Accurate Are We?
park.fan uses its own AI model trained on millions of real queue records to forecast crowd levels and wait times at theme parks worldwide. The model factors in weather forecasts, school holiday calendars, and known events — and retrains automatically as fresh data arrives.
How Predictions Are Made
Every wait time forecast is built from three layers of data: historical queue records collected from parks over years, external signals like weather forecasts and school holiday calendars, and contextual factors such as special events and seasonal patterns. The model generates both hourly wait time forecasts for individual rides and daily crowd level ratings for full park visits — available up to 365 days ahead.
Actively Developed — Improving Every Day
park.fan is a living platform, still in active development. Every day the system automatically compares its earlier predictions against the wait times actually recorded at the parks, feeding those comparisons back into the next training cycle. The accuracy metrics above reflect this real-world performance — not synthetic benchmarks. New parks and attractions are added regularly, and the model becomes more accurate as it accumulates more verified data for each destination.
Current edition
v20260720_0600
Trained
Jul 20, 2026
Training basis
2.2M data points · over 208 days
Where Fancast has been most on-point lately
The rides whose recent forecasts landed closest to the real wait — average error in minutes, live from the model.
| Ride | Park | Avg error |
|---|---|---|
| Bumper CarEverland | ±2.4 min | |
| Shooting GhostEverland | ±2.4 min | |
| Bow Wow BlasterSchlitterbahn NB | ±2.4 min | |
| JukeBoxLiseberg | ±2.8 min | |
| HurricaneEverland | ±3.2 min |
02
The ingredients
What Fancast reads
A rainy bridge-day in October is a completely different animal from a sunny holiday Saturday in July — and a model has to learn that first. So Fancast feeds on several sources at once:
Live wait times
Millions of real readings from 150+ parks, updated by the minute. The raw currency of every forecast.
Calendars & holidays
Weekends, public holidays and school breaks — including neighbouring regions, because day-trippers do not care about borders.
Weather
Rain probability and temperature bend the near-term forecasts. Sun pulls crowds in, all-day rain empties the paths.
Events & season
Halloween, summer holidays, long weekends, a headliner in its first summer — the usual suspects for a packed day.
History
Years of wait-time history per park. Patterns you only see if you stare at them long enough.
Hours & capacity
When the park opens, for how long, at what capacity — the frame everything else fits into.
Out of this mix the model makes two things: an hourly wait-time forecast for individual rides and a daily crowd-level grade for the whole park.
03
At real parks
Fancast at three parks
All theory is grey — Fancast only gets tangible at an actual park. Three examples of how the same ingredients turn into three completely different forecasts:
Europa-Park · bridge-day in October
Calm, green, under 30 minutes
Fancast sees: school holidays in only one neighbouring region, mixed weather, no special event. Result: a calm, green forecast — Voltron Nevera probably under 30 minutes, blue fire a walk-on. The same park three weeks later on a holiday Saturday? Deep red. Six million yearly guests do not spread themselves out on their own.
Very Low
Phantasialand · holiday Saturday
Compact, packed, orange to red
Compact park, few headliners, everyone wants Taron — saturation arrives faster than the first beer is poured. Fancast knows this and paints the day orange to red. The calendar next to it promptly suggests the Tuesday after, when you can ride Taron back-to-back instead of just longing for it.
Very High
Efteling · rainy Tuesday in November
The insider tip the model already counts in
Exactly the day gut-feeling planners avoid — and that Fancast paints green. Few holidays, miserable weather, short queues. It works precisely until everyone has read the same insider tip; which is why the model folds the rain probability in itself, instead of relying on folklore.
Low
04
The method
How Fancast learns (and cannot cheat)
The most important trick is an unglamorous one: Fancast retrains every night, every day at 06:00 UTC. Whatever happened yesterday, the model knows today. A coaster fan gets older and more tired over the years — Fancast gets a little smarter every morning.
And it is only ever tested on days it has never seen — on the future, not on memorised days from the past. Anything else would be like slipping yourself the exam questions in advance and then celebrating your straight-A report card.
On top of that, Fancast watches whether it is drifting — whether reality is slowly running away from it. And a new model version only goes live if it genuinely beats the old one in a fair head-to-head. Democracy among algorithms: if you are not better, you stay on the bench.
05
The scale
Green, yellow, red: the crowd levels
At the end of all that arithmetic sits a single colour. Six levels, from “you have basically got the park to yourself” to “welcome to a holiday Saturday”:
Very Low
Almost empty. Rope-drop dreams, back-to-back rides, a photo with the mascot and no queue.
Low
Relaxed. Short waits, you get on everything without needing a battle plan.
Normal
Normal operation. The headliners fill up, the rest stays easy-going. A solid compromise day.
High
Noticeably busy. For the top rides it pays to get up early — or to bring patience.
Very High
Properly busy. Long queues at the highlights; planning clearly beats spontaneity.
Extreme
Full alert. Holiday Saturday in high summer. Only with a strategy, stamina and a sense of humour.
06
Try it yourself
Grab a park
Enough theory. Fancast runs on every park page — here are a few popular ones to try it on directly. Click in, open the crowd calendar, and see which colour your chosen day gets:
07
All over the park
Where you meet Fancast
Fancast does not live on one lonely page — it is woven through all of park.fan, usually without introducing itself:
Today’s forecast
the crowd-level grade in the park header, before you even tap the first ride.
Crowd calendar
the calendar of best days to visit on every park page — green, yellow, red, up to a year ahead.
Best time to visit
the quietest weekdays and the upcoming insider days, distilled from the same data.
AI forecast in the wait-time chart
the dashed line that reveals a ride’s cheapest time windows.
Rope-drop recommendation
the honest answer to “is it worth arriving early?” — including the expected troughs.
No forecast
honest over guessed: parks with too little data get No forecast instead of an invented number.
How it all plays out inside a park is walked through step by step in the full guide — crowd calendar, badges and live wait times included.
08
In brief
Frequently asked about Fancast
How accurate is Fancast?
The current accuracy is shown live in the scorecard above — as MAE (average error in minutes), RMSE and MAPE. Those figures come from actually comparing past predictions with the wait times that were really measured, not from a flattering test lab. They change whenever the model retrains.
How far ahead can Fancast predict?
Fancast gives daily crowd levels for a park up to 365 days ahead. For individual rides it also produces hourly wait-time forecasts. The closer the day gets, the more short-term signals like the weather forecast are factored in.
How does Fancast know a holiday Saturday will be busy?
From the interplay of many signals: school and public holiday calendars (including neighbouring regions), the day of week, the weather forecast, special events and the park’s full wait-time history. A holiday Saturday in high summer carries almost all of those factors at once — which is why the forecast spikes there, while a rainy Tuesday in November stays green.
How often is the model updated?
Every day. Fancast automatically retrains once a day at 06:00 UTC on the freshest data — including yesterday’s wait times. So it literally gets a little better every morning.
Can I use Fancast for a specific park and day?
Yes. Every park page on park.fan has a crowd calendar that shows you a green, yellow or red forecast for each individual day up to a year ahead — from Europa-Park to Phantasialand, Efteling and Walt Disney World. You also get hourly wait-time forecasts for the individual rides.
What data does Fancast use?
Live and historical wait times from over 150 parks, school and public holiday calendars (including neighbouring regions), weather forecasts, opening hours, special events and seasonal patterns. That mix produces the daily crowd levels and the hourly wait-time forecasts.
Why does a park show “No forecast”?
Fancast only rates a park once there is enough operating data — at least around 30 operating days. Brand-new or rarely-open parks do not have that basis yet, so we would rather honestly show “No forecast” than a guessed number.
Does Fancast cost anything?
No. Like all of park.fan, every forecast, crowd calendar and statistic is free, ad-free and usable without an account.
