0
$\begingroup$

I tried making a model using the autoTS library but the thing is in the result it gives me the following results. I checked everything there is no missing data but the original data had a missing month so I added that and did quantity to zero but still when I see the results for actual vs predicted or RMSE value of models it doesn't go the right way.

Here are the images: This is for training and testing splitting with results after adding missing months enter image description here

enter image description here

This is for making the model using AutoTS enter image description here

This the result

enter image description here

This the result for ML

enter image description here

$\endgroup$

1 Answer 1

2
$\begingroup$

First, your absolute values, for the time-series are extremely large. Quite often these packages would then try to compute variance, and you get numerical overflows. Consider normalizing this.

Second, series is clearly not suited for ARIMA-type things (I would guess this includes VAR) since values of your series does not average to zero, and there is a hard-minimum.

Third, without any additional variables, I would be surprised if you got a decent prediction. Peaks seem quite random and you only have few datapoints. Without some very deep knowledge of the system, or some very predictive variables, or very long time series, I would expect prediction to be little different from running average.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.