Naive Forecast

The Naive Forecast method is one of the simplest forecasting techniques which assumes that the most recent observation is the only important one and forecasts the same value for all future points.

Simple Explanation

Imagine you are trying to predict the number of apples you'll sell tomorrow. A Naive Forecast would say: "However many you sold today, that's how many you'll sell tomorrow." It's quick, straightforward, and sometimes surprisingly effective as a baseline.

Formula

The formula for the Naive Forecast for a time series is simply:

ŷₜ₊₁ = yₜ

where ŷₜ₊₁ is the forecast for the next period, and yₜ is the actual observation at the current time period.

Parameters

There are no parameters to tune in a Naive Forecast model, which is part of what makes it so simple.

Feature Importance

Feature importance does not apply to Naive Forecast because it does not use any features or variables other than the past value of the series itself.

The Naive Forecast is often used as a benchmark for more complex forecasting models. If a sophisticated model can't beat the Naive Forecast, it might be too complex or not suitable for the data at hand.

Code

To see how we implemented Naive Forecast in our project NOT including cross-validation, check out the Naive Forecast notebook.