Built on public data

Predicting seasonal demand: a forecasting write-up

A demonstration forecasting model on public data, documenting method, accuracy in the 82-89% range, and where the model should not be trusted.

Demonstration project

This analysis uses publicly available data, not client data. It exists to show method and standard of work.

The question

How far ahead can a small service business plan staffing before the forecast stops being useful?

Public weather, seasonal employment and service demand data for the Gulf Coast region.

A classification and regression workflow in Python, validated out of sample, with documented assumptions, error rates and failure modes.

Predicted versus actual demand by week

Accuracy landed in the 82–89% range across validation folds for four-to-six week horizons.
Beyond eight weeks, accuracy decayed to a point where a simple seasonal average performed nearly as well.
Storm events are the dominant failure mode; the model is explicitly not reliable across them.

Tell us what you are trying to figure out. If we are not the right fit, we will say so on the call.