Table 1 Performance comparison of vulnerability detection methods across different vulnerability types.

From: A graph attention network-based multi-agent reinforcement learning framework for robust detection of smart contract vulnerabilities

Methods

Reentrancy

Unchecked Low Level Calls

Front Running

Denial of Service

Acc

Rec

Prec

F1

Acc

Rec

Prec

F1

Acc

Rec

Prec

F1

Acc

Rec

Prec

F1

Smartcheck

62.5

58.3

55.7

57.0

58.2

54.1

52.3

53.2

55.8

51.2

49.6

50.4

60.1

56.4

53.8

55.1

Mythril

75.3

71.8

68.9

70.3

72.4

68.7

65.2

66.9

69.5

65.8

62.4

64.1

71.8

68.2

64.7

66.4

Slither

78.9

75.2

72.6

73.9

76.1

72.4

69.1

70.7

73.2

69.5

66.3

67.9

75.5

71.8

68.4

70.1

MythSlith

82.4

78.9

76.2

77.5

79.8

76.1

73.2

74.6

76.9

73.2

70.1

71.6

79.2

75.6

72.3

73.9

GNN

85.7

82.3

79.8

81.0

83.2

79.8

76.9

78.3

80.4

76.8

73.9

75.3

82.7

79.2

76.1

77.6

GCN

87.9

84.6

82.1

83.3

85.6

82.3

79.5

80.9

82.8

79.4

76.5

77.9

85.1

81.7

78.8

80.2

DR-GCN

89.2

86.1

83.7

84.9

87.1

83.9

81.2

82.5

84.5

81.2

78.4

79.8

86.8

83.4

80.6

82.0

DA-GNN

91.5

88.4

86.2

87.3

89.3

86.1

83.6

84.8

86.7

83.5

80.8

82.1

88.9

85.6

82.9

84.2

GANs-MARL

93.8

90.7

88.9

89.8

91.6

88.4

86.1

87.2

88.9

85.8

83.2

84.5

91.2

88.1

85.4

86.7

  1. Values are shown in percentages (%). Acc = Accuracy, Rec = Recall, Prec = Precision. Best results are shown in bold.