# Forecasting

> Using past data to estimate future numbers, with an honest range around the guess.

- Level: Intermediate
- Topics: Data, AI
- URL: https://tidbit.legios.com.ar/forecasting/
- In Spanish: [Pronóstico](https://tidbit.legios.com.ar/es/forecasting/)

**Forecasting** is estimating future values from past data: how many umbrellas a store will sell next month, how much electricity a city will use tomorrow at 7 PM, how many calls a support center will get on Monday. Most forecasts start from a time series, a sequence of measurements over time.

A forecasting model looks for patterns it expects to continue. **Trend** is the long-run direction, up or down. **Seasonality** is the repeating rhythm, like higher sales every December or lower traffic every Sunday. Classic methods such as moving averages, exponential smoothing and ARIMA capture these; machine learning models can add outside signals like weather, prices or holidays.

A good forecast is not one number but a range: "between 900 and 1,100 units, most likely around 1,000". The further ahead you look, the wider that range gets. Teams check models by **backtesting**: pretending it is last year, forecasting, and comparing with what actually happened. Forecasts drive real decisions, such as how much stock to order or how many people to schedule, so being clear about the uncertainty matters as much as the number itself.

## An analogy

Like a weather forecast: it reads patterns from the past to say "70% chance of rain tomorrow", is quite reliable for the next day, and gets fuzzier the further out you look.

## Keep in mind

- Estimates future values from historical data, usually a time series.
- Looks for trend and seasonality; can add outside factors like weather or holidays.
- A useful forecast comes with a range, which widens the further ahead it goes.
- Backtesting against past periods shows how much to trust a model.

## Go deeper

- [Forecasting on Wikipedia](https://en.wikipedia.org/wiki/Forecasting)
- [What is forecasting? (IBM)](https://www.ibm.com/think/topics/forecasting)

## Related terms

- [Time series](https://tidbit.legios.com.ar/time-series/): A list of measurements of the same thing, taken one after another in time.
- [Machine learning](https://tidbit.legios.com.ar/machine-learning/): Teaching computers by showing them examples instead of writing every rule.