[ad_1]
Utilizing a number of mannequin types to seize and forecast the elements of complicated time sequence
I not too long ago needed to repair the fence in my again yard. It’s outdated, wood, and has been threatening to topple over for some time now. Between curses it actually struck me what number of instruments I wanted to make use of to get the job executed, and the way typically you really want multiple software for the job.
What does this need to do with time sequence regression? On the whole, little or no. Particularly, fairly a bit: at this time we’ll be diving into utilizing combined fashions for time sequence evaluation and forecasting. Or in additional DIY phrases — utilizing extra mannequin instruments to get the forecasting job executed.
So, with out an excessive amount of extra rambling, we are going to get cracking with:
- Revisiting the large image, and speaking a bit about combined fashions.
- Taking a look at some real-world knowledge.
- Utilizing a easy mannequin to seize the pattern in our time sequence.
- Seasonality 3 ways: resolution bushes, linear regression, and traditional time sequence.
- Placing Humpty Dumpty again collectively to get a single time sequence prediction.
Apart: you’ll be happy to know that my thumbs have made a full restoration following some hammer-related accidents.
As at all times, we’re attempting to construct probably the most “correct” mannequin potential. On this case, we’re specializing in forecasting so we’ll prioritise fashions which might produce probably the most correct estimates of future time sequence values.
I’ve beforehand written at size about utilizing a regression method primarily based on Meta’s Prophet methodology. In these articles, I defined why I selected to make use of the LASSO mannequin: non-stationarity made utilizing traditional strategies tough, the necessity for wise extrapolation dominated out tree-based approaches, and the will for simplicity and explainability excluded any type of neural community.
However I additionally hinted at utilizing combined mannequin types — that’s, modelling every of the time sequence elements utilizing a unique mannequin type, and mixing the output of the…
[ad_2]
