Antillia
PLLet’s talk ↗

APPLIED INTELLIGENCE / Forecasting

See what demand history can tell you.

A 24-hour forecast compared with what actually happened.

Local experiment case study

Compare forecasts ↓
DemandCast — recorded local application interface

Recorded local application view. This image is not an interactive interface.

EXPERIMENT EXPLORER

Forecast meets reality.

Choose a recorded window and compare models.
Historical results; no new training.

900450008:0016:0000:0007:00
Model forecastObserved demand

2012-09-13 08:00:00 · Seasonal baseline

View values in a table
Horizon hourForecastObserved demand
1646.00758
2123.00338
349.00212
462.00254
5137.00297
6191.00283
7216.00274
8270.00352
9416.00491
10808.00884
11835.00852
12558.00674
13406.00463
14272.00317
15217.00251
16169.00170
1796.0087
1829.0041
1911.0030
209.0014
218.0012
2240.0041
23140.00152
24486.00454

Source: UCI Bike Sharing, CC BY 4.0. Data and forecasts from the published project report. MAE above describes the selected window, not the full test.

RESULT AND CONTEXT

What was tested.

66.01MAE · rentals / h
24 hForecast horizon

Method

168 hours of history and known calendar features. The seasonal baseline beat boosting and GRU on validation.

168 hours of history→24 forecast hours

Where this result stops

One city, 2011–2012 data. This is a historical experiment, not a forecast of current demand.

Results come from the project reports. This presentation does not include retraining or independent evaluation.

Download results ↓Method and code on GitHub ↗

YOUR NEXT PROJECT

What could work
better for you?

Let’s talk about your project ↗