Forecasting is one of the most important and frequent problems in machine learning.
- How many customers will visit a store next Friday?
- How many products will a factory sell three months from now?
- How many tourists will visit a city next summer?
In this talk, you will learn how to solve these types of problems with machine learning in Python.
The talk will start with a brief introduction to the necessary machine learning concepts, then explain how to transform a time series problem into a machine learning problem. It will focus on concepts that are not only essential for this approach, but for time series modelling in general.
Time series problems can be tackled with specialized models, but in practice, many real-world solutions rely on reframing forecasting tasks as supervised machine learning problems. In this talk, we’ll focus on this approach, since many of its core ideas also underpin time series–specific models. This makes it the best introduction to time series modelling.
We’ll start by reviewing the essential supervised machine learning concepts needed when working with time-dependent data. From there, we’ll see how to transform a time series problem into a supervised machine learning problem by carefully designing inputs and targets.
Finally, we’ll focus on a critical aspect: evaluation. We’ll discuss how to validate machine learning models in general, and how time dependence fundamentally changes validation strategies in the context of time series.
By the end of the talk, attendees familiar with machine learning will know how to apply their skills to time series problems, while those newer to ML will gain a solid conceptual foundation for understanding how modern time series models work under the hood.
Machine learning