Table 6 Comparison of model performance on the test set with and without data augmentation.

Model	Condition	Key optimized hyperparameters	CV performance (on development set)					Final performance (on hold-out test set)
Model	Condition	Key optimized hyperparameters	CV train MAE	CV train R²	CV valid MAE	CV valid RMSE	CV valid R²	Test MAE	Test RMSE	Test R²
XGBoost	Baseline	n_estimators = 500, max_depth = 5, learning_rate = 0.05	68.66	0.98	587.65	742.96	0.42	642.40	751.26	0.48
XGBoost	Augmented	n_estimators = 400, max_depth = 3, learning_rate = 0.05	276.98	0.84	614.90	759.11	0.39	632.05	715.97	0.53
SVR	Baseline	kernel=’poly’, C = 500, degree = 3	356.48	0.65	644.75	830.19	0.24	620.12	773.31	0.45
SVR	Augmented	kernel=’rbf’, C = 500, gamma=’auto’	284.18	0.73	660.34	840.97	0.26	644.96	765.48	0.46
MLP	Baseline	hidden_dims=[32, 16], learning_rate = 0.001, batch_size = 16	434.94	0.71	638.54	785.69	0.33	556.96	686.76	0.57
MLP	Augmented	hidden_dims=[128, 64], learning_rate = 0.005, batch_size = 16	256.23	0.86	706.37	868.69	0.20	641.48	771.49	0.46

Note: Bold values highlight the metrics of the best-performing model on the hold-out test set for each data condition (Baseline and Augmented).

Quick links

Search