Machine Learning (ML) · Advanced level

Advanced level · Module 3: Time Series and Forecasting · Lesson 2 of 4

Baselines and Feature-Based Forecasting

Start with naive baselines, then use lag and rolling features with regression models.

In this lesson you will

  • Build naive and seasonal-naive forecasts
  • Create lag and rolling features for ML forecasting

Key concepts

naive forecastseasonal-naive forecastlag featuresrolling statistics

The lesson

A naive forecast predicts next month equals this month; a seasonal-naive forecast predicts it equals the same month last year. They are quick, honest benchmarks.

To use ML models, turn the series into a table: each row is a month, with features such as last month's price (lag 1), the price twelve months ago (lag 12), recent averages and calendar month.

Create every feature only from information available at forecast time. Using the current month's average to predict the current month is leakage.

Step by step

  1. Compute naive and seasonal-naive forecasts.
  2. Build lag and rolling features.
  3. Train a gradient boosting model on them.
  4. Compare errors with the baselines.

Practice activity

Build baseline and feature-based price forecasts for the four markets.

Expected result: Forecasts for each market with errors compared against both baselines.

Check your understanding

  1. What does a seasonal-naive forecast predict?

    Show answer

    The value from the same period in the previous season, such as last year's same month.

  2. What is a lag feature?

    Show answer

    A past value of the series used as an input, such as last month's price.

Taught live by an instructor — online or in person. Free course lessons are for self-study and do not lead to a certificate — see registration options.