ARIMA Forecasting in R

[@georgew7814]

are you incorporating the auto.arima() function in the forecast package into your modeltime package?

[@BusinessScience]

Yes, the arima_reg() model spec connects to forecast::Arima() and forecast::auto.arima() depending on the engine you set.

[@drasko40]

Could you explain what is nesting_column? Or is it that "nest" is a function and a nesting_column is predefined? Thanks

[@BusinessScience]

Hey, nested columns are just columns that contain a complex data structure stored inside of lists. Typically data frames, which is what modeltime stores.

[@Miyazaki97]

How can I include modeltime_calibrate to get the confidence interval in the nested model_table?

[@BusinessScience]

This is pretty easy to do. When you calibrate, just use modeltime_forecast() and you can get the confidence intervals.

[@Miyazaki97]

Thank you for your kind reply. This is what I did but it didn't work. mutate(nested_forecast = map2(fitted_model, nested_column, .f = function(arima_model, df){ modeltime_table( arima_model ) %>% modeltime_calibrate(df)%>% modeltime_forecast( h = 30, conf_interval = 0.95, actual_data = df, keep_data = T) I am sorry I am not familiar with nesting. Could you please help me?

[@BusinessScience]

I think the map and nested structure is throwing you off. Simply the problem. Just do one without the nest - see if you can get the CI. Then turn it into a function that you can map.

[@BusinessScience]

One more thing - the docs exist to help you. Qbusiness-science.github.io/modeltime/

[@Miyazaki97]

Thank you for your enormous support and guidance.

[@alvaromorales6828]

How do you deal with missing values?

[@BusinessScience]

There are several ways to deal with missing values. The timetk package that I’ve created has padding and imputation capabilities. business-science.github.io/timetk/

[@kunalsatpute8379]

what was the accuracy of the model?

[@BusinessScience]

There were 7 models. The accuracies can be obtained by splitting each iteration and evaluating with modeltime_accuracy(). This is just a quick R-Tip, so I didn't go into the details there. Also, the accuracy can be vastly improved. We're using pretty simple techniques (ARIMA) here, but machine learning offers a BIG improvement. With our Nostradamus Auto-Forecasting app, we were getting around 5000 RMSE on this data. Simple ARIMA is probably on the order of 10,000+ (much worse).

[@sulochandhungel]

Links?

[@BusinessScience]

Which links do you need?

[@sulochandhungel]

@@BusinessScience the video says there are links on the description. I also wanted to see if this course would be helpful to understand stochastic hydrology and time series. I'm having trouble finding a r based course on that.

[@BusinessScience]

@@sulochandhungel Sorry - Just added the links. You can check out the R-Track here. university.business-science.io/p/5-course-bundle-machine-learning-web-apps-time-series/

[@BusinessScience]

Time Series Course Here: university.business-science.io/p/ds4b-203-r-high-performance-time-series-forecasting

[@maksim0933]

why forecast on plot 1_38 insensitive to wavering? Just a straight line

[@BusinessScience]

That’s the downside of auto arima. You need to use better methods if you expect better results.

[@maksim0933]

@@BusinessScience thank you for your reply, and one more question: is it possible to combine your method with prophet package? I mean, instead of auto.arima from forecast package use prophet?

[@BusinessScience]

Yes - I teach ARIMA, Prophet, Prophet Boost (my invention), Machine Learning & Deep Learning in my Time Series Course. university.business-science.io/p/ds4b-203-r-high-performance-time-series-forecasting

[@janiobachmann5029]

Also, this is due because the model was not able to capture seasonality. When Matt talks about more advanced concepts one that will definitely help your models is the concept of "Feature Engineering", this will better help your model capture seasonality. These concepts I learned in the time series course of Matt (Note I am not a sponsor), I am just a student of Matt who enjoys taking his courses!