Replacing last line missing value for time series

Helios

Hi everyone,

I just want to do a simple prediction for our company sale numbers. I have data from 2012 to 2020 for one of our product and i just want to predict 2021 year sale number so i can compare it with the real sale numbers.

I just do not want to do it with replacing missing numbers with average or other methods. Tried to do it with "replacing missing numbers (time series)" operator with linear interpolation but it seems it can not replace the last line of missing value. How can i solve this or should i use knn, decision tree or neural network to predict it?

If we manage to predict number i also want to compare the methods' prediction values. i would appreciate if it is a simple solution like (hard to understand and explain root mean squared error to boss for example) comparing the prediction values how close to real numbers.

Maybe i did all the process wrong but ty for your answers i am still trying to learn.

ceaperez

Hi @Helios, 

I suggest you two simple options

the first one is to check the "ensure finite values"  option into the Replace missing values operator. 

the second one is to use a simple Arima model to predict the last missing value. 

please find attached the modified process.

Best, 

Cesar

Helios

Hey @ceaperez ty for your answer,

I checked the process you sent and really appreciate for that.

For the "replace missing values (time series)" operator, even i checked the ensure infinite values option, it still replacing the missing value like last value. so it still not working as linear interpolation or i'm doing something wrong.



Arima model worked really fine. i got this result:

But i worked on that ARIMA model, after i deselect "estimate constant" option i got closer result to real value which is 18.085.

I don't know disabling estimate constant option is a good decision or not but i'm still searching for better or optional models. but really huge thanks to @ceaperez