1

I am looking for some advice and guidance - I'm completing a Lvl 4 apprenticeship and I need to complete a time series forecast.

Attempting to predict a target number of reports (x) on a category by category basis utilising time series. I have grouped by data by category and by time, however, when processing in Orange, the ARIMA is showing too few values, which is correct. What is the best way to time series forecast using a small dataset?

I have tried VAR, however, this is stating that my target var needs to be categorical... I'm looking for a numeric target

Basically, i'm trying to forecast that at x location, in the next few hours, there will be y reports... then using that list of x and y, impose that data on a GIS heatmap

My SQL Query (pseudo) is: SELECT Location, CountOfReports,TImeOfDay FROM ListOfReports WHERE DayOfWeek = Today.DayOfWeek GROUP BY DayOfWeek, TimeOfDay

This gives me a summary breakdown, but is resulting in too few records.

Am I approaching this wrong?

Shaun R
  • 11
  • 1
  • The duplicate should be helpful. That said, "the ARIMA is showing too few values" sounds strange, because in principle, you should be able to *fit* an ARIMA to very little data (it's just that it won't be any good). Also, VAR (vector autoregression) is inherently numerical, so it's again strange that your tool wants categorical data. I recommend this textbook: [*Forecasting: Principles and Practice* (3rd ed.) by Athanasopoulos & Hyndman](https://otexts.org/fpp3/). When you write about locations, you may want to think about spatial statistics, since nearby locations will likely be correlated. – Stephan Kolassa Sep 22 '21 at 06:23

0 Answers0