The Best Deep Learning Models for Time Series Forecasting
Everything you need to know about Time Series and Deep Learning
I’ve launched AI Horizon Forecast, a newsletter focusing on time-series and innovative AI research. Subscribe here to broaden your horizons!
Preliminaries
The landscape of Time Series forecasting has changed dramatically in the span of two years.
The forth and fifth in the series of Makridakis M-competitions (better known as M4 and M5 competitions respectively) took place in 2018 and 2020. For those who are not aware, these M-competitions are essentially a status-quo for the time series ecosystem, offering empirical and objective evidence that guides the theory and practice of forecasting.
The results of the M4 competition in 2018 showed that the pure ‘ML’ methods were largely outperformed by the traditional statistical approaches. This was unexpected, given that Deep Learning had already left an indelible imprint on other fields such as Computer Vision and NLP. However, in the M5 competition[1] two years later, with a more creative dataset, the top spot submissions featured only ‘ML’ methods. To be more precise, all 50 top-performing methods were ML-based. This competition saw the rise of the versatile LightGBM (used for time…
