Table 3 Model performance metrics for different validation strategies.

From: Global high-resolution ultrafine particle number concentrations through data fusion with machine learning

Method

MAE

MSE

R2

Perc. Error

10% Test set

1,336

2,644

0.90

23%

10-fold CV

1,025

1,541

0.91

24%

Spatial LOOCV

2,839

9,427

0.77

32%

Temporal LOOCV

1,740

6,912

0.87

26%

  1. MAE and MSE are given in cm−3. The percentage error represents the mean relative error across all predictions. Results show performance for: traditional test set evaluation (10% of data), 10-fold cross-validation, spatial Leave-One-Out Cross-Validation (LOOCV) where individual stations are held out, and temporal LOOCV where entire years are held out.
Back to article page