Improving transaction safety via anti-fraud protection based on blockchain

Figures & data

Figure 1. Linked list structure of blockchain.

Figure 2. The structure of binary Merkel tree.

Figure 3. An overview of GBDT-APBT where multiple experts voted, followed by a consensus model that determines whether it is a fraud transaction.

Figure 4. Transfer process.

Figure 5. Credit card fraud detection framework.

Table 1. Results of the vote.

Model	$f_t^{(a)}$	$f_t^{(b)}$	$f_t^{(c)}$	$f_t^{(d)}$	$f_t^{(e)}$
Node a	1	0	0	1	1
Node b	1	1	0	1	0
Node c	0	0	1	1	1
Node d	1	1	0	1	0
Node e	1	0	0	1	1
Accuracy($\mathrm{\%}$)	80	40	20	100	60

Figure 6. The distribution of data categories.

Figure 7. The distribution of transaction amount.

Figure 8. Confusion matrix of different algorithm models.

Figure 9. Precision-Recall curve of different algorithm models.

Table 2. The result of predictions.

	TN	FP	FN	TP	$R=\frac{TP}{TP+FN}$(%)
GBDT	85,008	139	36	112	75.68
Bayes	85,027	120	48	100	67.57
DT	78,960	6187	49	99	66.89
SVM	85,055	92	25	123	83.11
KNN	76,684	8463	52	96	64.86
XGBoost	85,279	16	37	111	75.00
RF	85,300	7	26	110	80.88
CNN	85,278	24	30	111	78.72
DNN	85,278	18	31	116	78.91
GBDT-APBT	85,069	78	10	138	93.24

Table 3. The result of predictions.

	AUC-ROC performance (%)	AUC-PR performance (%)
DevNet	97.85	75.60
REPEN	97.20	69.40
DSVDD	97.70	72.80
DIF	95.30	48.70
LeSiNN	96.28	63.10
GBDT-APBT	98.20	87.10

Table 4. Heart failure prediction dataset.

	Precision (%)	Recall (%)	F1-score (%)
GBDT	84.30	85.80	85.04
Bayes	86.40	84.80	85.07
DT	79.80	73.00	75.65
SVM	84.80	84.00	83.95
KNN	85.80	86.60	85.97
XGBoost	83.60	81.00	81.61
RF	85.20	84.80	84.43
GBDT-APBT	91.50	88.30	89.87

Table 5. Default of credit card clients dataset.

	Precision (%)	Recall (%)	F1-score (%)
GBDT	87.60	83.50	85.50
Bayes	88.10	86.10	87.03
DT	86.00	78.50	81.80
SVM	83.30	84.00	83.65
KNN	79.00	72.50	75.61
XGBoost	79.00	88.30	83.39
RF	84.00	88.20	86.05
GBDT-APBT	88.60	89.60	89.10