Time Series forecasting with multiple series with irregular frequency

Question

I had been attracted to DeepAR for forecasting on a combination of many series.

However, my series data is sales data, and the frequency is irregular, even within each series. It seems that within unsampling the data – which is highly undesirable – this might not be a suitable model.

I would be content if this is not a model that exists at the moment to build a prediction model for each series individually, as long as that did not involve unsampling or interpolation.

What advice would people have around this? Is there a model that supports many irregular series? Or, failing that, which model would be most suited to predicting on an irregular time series?

Answer

Hi, you may want to follow this Kaggle training: "Forecasting With Machine Learning"
See https://www.kaggle.com/code/ryanholbrook/forecasting-with-machine-learning/data

```
you will predict sales for the thousands of product families sold at Favorita
 stores located in Ecuador. The training data includes dates, store and 
product information, whether that item was being promoted, as well as 
the sales numbers. Additional files include supplementary information that
 may be useful in building your models.
```

Answer

Awesome, thanks. I’ll have a look.

Time Series forecasting with multiple series with irregular frequency

関連するコンテンツ