PROJECT / DEMAND-FORECASTING
Forecasting demand.
An app for comparing actual bike rental demand with a model's estimates, and exploring its predictions under different weather conditions.
Reproducible projectWhat you can explore
The chart brings together three lines: actual rentals, the model's prediction and a simple estimate based on the same hour of the previous week. Browse different periods to see where the model gets close and where it misses. Then choose an observation and change its weather conditions to compare the original prediction with the new scenario. The app is built in Python with Streamlit and uses UCI's public bike rental data.
What the project tells us
On the period reserved for evaluation, the model's mean absolute error was about 45 rentals per hour, compared with 64 for the weekly reference. The charts reveal something the average alone cannot: when those errors happen. This is a one-hour-ahead exercise using observed history and assuming that hour's weather is known, not a forecast for several days. Changing the weather in the demo shows how the model responds; it does not prove that a weather change causes a specific change in demand.
EXECUTED EVALUATION / 2026.09.30
First 168 observations of the historical test. Rentals per hour; this chart is not a live forecast.