Table 5 Impact of training under imbalanced class distribution (1:5 ratio).

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

Metric

Original training (Balanced)

Imbalanced (1:5)

+ Weighted Loss

\(\Delta\) (Weighted vs. Imbalanced)

AUC (%)

91.3

88.7

89.4

+0.7

Recall (%)

85.9

78.2

82.4

+4.2

Precision (%)

87.2

85.6

84.9

-0.7

F1-Score (%)

86.5

81.8

83.6

+1.8

Back to article page