Search code examples
time-seriesxgboostmetrics

Time series with XGBoost

On some time series data I am working with XGBoost and I am getting a large value of RMSE:

enter image description here

I scaled all the data (including the target) and I got the logic results of values between 0 and 1:

enter image description here

I'm not sure if I can say that my model is accurate according to the scaled data values?

Solution

  • Generally, we use MAE as the test statistic for real-world data.

    High MSE is an indicator that there are big outliers in your predictions.

    MAE vs MSE:

    Mean Absolute Error (MAE) is less susceptible to outliers since it does not "penalise" outliers.
    It is used in cases where performance is measured using continuous variable data.
    It produces a linear number that equalizes the weighted individual disparities.

    Mean Squared Error (MSE) is more susceptible to outliers as it "penalise" outliers heavily.
    This metric excels when the dataset contains outliers, or unexpected values (too high or too low values).

    Additional tips:

    You should also look into the Root Mean Squared Error (RMSE) metric.
    This allows you to identify your model prediction errors to fix it.
    If RMSE is close to MAE, the model makes many relatively small errors.
    If RMSE is close to MSE, the model makes few but large errors.
    MAE ≤ RMSE ≤ MSE (for Regression)