Long-Horizon Forecasting with Deep Learning: A Full Tutorial on TSMixer and iTransformer
How to use Deep Learning forecasting models the right way
In previous articles, we discussed TSMixer and iTransformer—2 milestone Deep Learning models.
Specifically, we explored their inner workings, strengths, and how they operate.
In short, both models leverage cross-variate information. Their feature-mixing components enable joint learning of interdependencies across covariates, modeling and predicting all dataset channels simultaneously.
Also, they excel in longer prediction lengths—scaling performance to at least 720 datapoints in benchmarks.
This article demonstrates how to use TSMixer and iTransformer to build an end-to-end Long-Horizon Forecasting project. We’ll also show how to tune these models and apply cross-validation for extra performance.
Let’s get started:
✅ Find the hands-on project for this article in the AI Projects folder (Project 8), along with other cool projects!
You can also support my work by subscribing to AI Horizon Forecast: