I have a time series technique I want to try out. Is there a list of canonical data sets or time series data that I should test my technique on so that I can see if the technique works? For example, when it comes to ML, the IRIS and MNIST are some popular datasets that one should test, and there are benchmarks to exceed.

In R, I could generate any order Arima() using Arima.sim() function. I plan to generate AR(1) and MA(1), probably ARMA(1, 1), AR(2), MA(2), and ARMA(2, 2), data. Are there any other ones I should generate to cover the main types of data? Maybe seasonal data or regression with time series error type data? And finally, are there any popular real-life time-series data sets I should test the technique on? The response can be either categorical or numerical.


2 Answers 2


I don't know any iconic MNIST-like data set for time series, but what I would suggest is to look for suitable real life data on the internet. Good sources are Kaggle or the UCI Machine Learning Repository.

Look for a few substantially different (different in the sense of from what scientific area the data originates) data sets where you have a strong baseline with current models / algorithms and compare them against your new technique.

  • Thanks. I will do that and probably look at some time-series textbooks and use some of the common data sets in it like the "airline" data set for seasonality.
    – confused
    Jun 22, 2020 at 9:30

The Monash Time Series Forecasting Repository contains multiple different time series datasets from various domains. These have been collected from forecasting competitions or other previous forecasting use, so you should also be able to learn something from previous work that used a particular dataset.

An introduction and description of the repository, along with more information about the datasets, is given by Godahewa et al. (2021).

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.