Time Series Forecasting Tutorial: Define Horizons and Evaluate Ordered Predictions

Time series forecasting predicts values beyond a defined observation time. Lagged features and seasonality describe temporal structure, while the forecast horizon determines which future values are required. Evaluation must preserve chronology and compare against suitable simple forecasts, because randomly mixing time points can reveal information unavailable when the forecast would have been issued.

Here's a popular Time Series Forecasting video — watch it, then summarize it with one click. Or paste any other YouTube link below.

— or paste your own YouTube link —

Want to keep the key points from a YouTube video? Paste the link below to turn it into structured notes.

Why Use Time Series Forecasting Tutorial: Define Horizons and Evaluate Ordered Predictions

Frequently Asked Questions

What is a forecast origin?

It is the time at which a forecast is issued using the information then available. Identifying it clarifies which measurements belong in the input and which future values form the target.

How does horizon change the problem?

Horizon specifies how far into the future predictions extend. Short and long horizons provide different information support and error behavior, so their results should not be combined without identifying which task was evaluated.

Why is random splitting often unsuitable for forecasting?

It can place future observations in training while evaluating earlier times. A chronological design better respects the sequence of information available when the intended forecast would actually be made.

What is a seasonal baseline?

It predicts using a corresponding earlier seasonal position, such as a previous recurring interval. Its suitability depends on the data's timing and provides a useful comparison for models claiming to capture temporal patterns.

Why evaluate several forecast origins?

Several origins expose behavior across different periods and conditions. One favorable interval can hide instability or changing relationships, so repeated backtesting provides a broader view of forecasting performance.