Machine Learning

How to Grid Search Naive Methods for Univariate Time Series Forecasting

Last Updated on February 27, 2020

Simple forecasting methods include naively using the last observation as the prediction or an average of prior observations.

It is important to evaluate the performance of simple forecasting methods on univariate time series forecasting problems before using more sophisticated methods as their performance provides a lower-bound and point of comparison that can be used to determine of a model has skill or not for a given problem.

Although simple, methods such as the naive and average forecast strategies can be tuned to a specific problem in terms of the choice of which prior observation to persist or how many prior observations to average. Often, tuning the hyperparameters of these simple strategies can provide a more robust and defensible lower bound on model performance, as well as surprising results that may inform the choice and configuration of more sophisticated methods.

In this tutorial, you will discover how to develop a framework from scratch for grid searching simple naive and averaging strategies for time series forecasting with univariate data.

After completing this tutorial, you will know: