Member-only story
Measure Performance for a Time Series Model [ETS or ARIMA]
Interpreting measures of error for time series data and understand the performance of your model to choose the best one
3 min readOct 30, 2020
In this article, we are going to talk about the types of error measuring techniques when dealing with the time-series data and how you can choose the best model from the ones you have created earlier such as ETS or the ARIMA model, and find the best ones based on the ACF and PACF plots.
Scale Dependent Errors
Scale-dependent errors, such as mean error (ME) mean percentage error (MPE), mean absolute error (MAE) and root mean squared error (RMSE), are based on a set scale, which for us is our time series, and cannot be used to make comparisons that are on a different scale. For example, we wouldn’t take these error values from a time series model of the sheep population in Scotland and compare it to corn production forecast in the United States.
- Mean Error (ME) shows the average of the difference between actual and forecasted values.
- Mean Percentage Error (MPE) shows the average of the percent difference between actual and forecasted values. Both the ME and MPE will help indicate whether the forecasts are biased to be…
