I built a model for time series in order to forecast new values.

What is the best way to choose the correct model?

Are better criteria like AIC and BIC or the comparison between prediction errors? In this last case I have to split dataset in train and test while in the first case is not necessary, reducing time of implementation.



First thing first, when ever you use Time Series data you call it as Forecasting not Prediction as it is time dependent. To understand why you can go through this link

Metrics to compare models

When you are trying to compare between models you need to use AIC,BIC, AUC etc values. you can go though this link to understand better

Metrics to Access the Model

When you are accessing the performance of the model then you need to check for Error Rates(RMSE, MAE, MAPE, MSE etc). Yes, in this case you need to divide the data into Train and Test to access the model. You can go through this link for better understanding

Improve the Forecast

To take it to the next level, you can use an ensemble to get better the result. This might or might not decrease the error rate. In most of the cases it is helpful. You can combine 2/3 moderately performing hotels outcome to get best results i.e., Ensemble Model.

| improve this answer | |
  • $\begingroup$ yes, 1st you need check the error rates and if they are satisfying enough based on the business scenario then you can choose the models which are good. Then you can compare between models using AIC, BIC, AUC which is best. If you want to take it even further use Ensemble and see how it performs. $\endgroup$ – Toros91 Apr 19 '18 at 8:24
  • $\begingroup$ So, If I understand well, to find a good result, I should follow these steps: 1) create many models and choose those with the lowest AICs/BICs 2) of these choosen models calculate the forecast error and choose 2/3 better models 3) ensemble them. Correct? $\endgroup$ – E.Wan Apr 19 '18 at 8:26
  • $\begingroup$ yes correct! this might help you in improving the results $\endgroup$ – Toros91 Apr 19 '18 at 8:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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