Table 2 Accuracy metrics for the best-performing ML model in nine experiments (k-fold cross-validation)

From: AI perceives like a local: predicting citizen deprivation perception using satellite imagery

ML experiments – Best performing (features)

max R2

min RMSE

mean R2

mean RMSE

sd R2

sd RMSE

A – SVM rad (LOG)

0.577

1.158

0.468

1.262

0.063

0.067

B – SVM rad (LOG)

0.636

1.072

0.555

1.154

0.044

0.052

C – RF (STD)

0.636

1.089

0.559

1.150

0.043

0.056

D – SVM rad (LOG)

0.665

1.025

0.601

1.092

0.045

0.065

E – SVM rad (STD)

0.211

1.575

0.104

1.659

0.047

0.092

F – SVM rad (STD)

0.221

1.507

0.152

1.614

0.035

0.096

G – RF (LOG)

0.601

1.090

0.556

1.156

0.034

0.042

H – SVM rad (STD)

0.695

0.953

0.647

1.03

0.033

0.045

I – RF (LOG)

0.707

0.972

0.667

1.010

0.027

0.046

  1. Each experiment (from A to I) involves a different set of input features, standardised (STD) or log-transformed (LOG), and the use of three types of ML algorithms (SVM, RF, XGBoost). In every experiment, the best-performing algorithm is either RF or SVM rad. The maximum (max), mean, and standard deviation (sd) of accuracy metrics (R2 and RMSE) are calculated on the validation fold.
Back to article page