Table 3 Results achieved using DXGBA and XGBoost with different sampling techniques (Bold values represent the highest performance).

From: Customs fraud detection using a gradient boosting approach for joint classification and risk estimation

Data	Model	Classification					Regression
Data	Model	Accuracy	Precision	Recall	F1-Score	AUC	MAE	MSE
Without SMOTE	XGBoost	0.8224	0.8446	0.2169	0.3452	0.8176	0.0008	0.000059
Without SMOTE	DXGBA	0.8244	0.8589	0.2227	0.3536	0.8212	0.0006	0.000053
SMOTE Balanced Resampling	XGBoost	0.8045	0.6330	0.2236	0.3301	0.7618	0.0008	0.000059
SMOTE Balanced Resampling	DXGBA	0.8053	0.6351	0.2306	0.3383	0.7692	0.0006	0.000052
Resample by 10% SMOTE	XGBoost	0.8197	0.7583	0.2411	0.3657	0.8012	0.0008	0.000059
Resample by 10% SMOTE	DXGBA	0.8210	0.7603	0.2489	0.3749	0.8050	0.0006	0.000053
Resample by 15% SMOTE	XGBoost	0.8112	0.7574	0.1846	0.2966	0.7835	0.0008	0.000059
Resample by 15% SMOTE	DXGBA	0.8130	0.7709	0.1903	0.3051	0.7904	0.0006	0.000053
Resample by 30% SMOTE	XGBoost	0.8113	0.7037	0.2172	0.3318	0.7841	0.0008	0.000059
Resample by 30% SMOTE	DXGBA	0.8137	0.7196	0.2243	0.3420	0.7859	0.0006	0.000052
Resample by 50% SMOTE	XGBoost	0.8077	0.6709	0.2134	0.3237	0.7703	0.0008	0.000059
Resample by 50% SMOTE	DXGBA	0.8111	0.6926	0.2249	0.3394	0.7786	0.0006	0.000053
RU	XGBoost	0.6629	0.3601	0.7234	0.4808	0.7558	0.0022	0.000057
RU	DXGBA	0.6673	0.3639	0.7242	0.4843	0.7608	0.0015	0.000051
Resample by 10% SMOTE + RU	XGBoost	0.7175	0.4007	0.6225	0.4875	0.7592	0.0007	0.000053
Resample by 10% SMOTE + RU	DXGBA	0.7194	0.4039	0.6305	0.4923	0.7638	0.0007	0.000049
Resample by 15% SMOTE + RU	XGBoost	0.7356	0.4169	0.5639	0.4793	0.7561	0.0008	0.000059
Resample by 15% SMOTE + RU	DXGBA	0.7387	0.4229	0.5774	0.4881	0.7619	0.0006	0.000054
Resample by 30% SMOTE + RU	XGBoost	0.7784	0.4847	0.4218	0.4508	0.7528	0.0008	0.000059
Resample by 30% SMOTE + RU	DXGBA	0.7817	0.4936	0.4305	0.4598	0.7624	0.0006	0.000053
Resample by 50% SMOTE + RU	XGBoost	0.8014	0.5765	0.2996	0.3943	0.7634	0.0009	0.000059
Resample by 50% SMOTE + RU	DXGBA	0.8030	0.5819	0.3108	0.4051	0.7675	0.0007	0.000052

Back to article page

Table 3 Results achieved using DXGBA and XGBoost with different sampling techniques (Bold values represent the highest performance).

Search

Quick links