Table 6 The performance of LLMs based-data augmentation on the NF-BoT-IoT-v2 dataset.

From: An IoT intrusion detection framework based on feature selection and large language models fine-tuning

 

Our

CTGAN

SMOTE

ADASYN

Oversample

Undersample

Original

Benign

0.955

0.956

0.934

0.885

0.934

0.836

0.952

DDoS

0.993

0.993

0.992

0.992

0.992

0.990

0.993

DoS

0.987

0.987

0.987

0.828

0.987

0.976

0.987

Theft

0.966

0.966

0.965

0.550

0.965

0.918

0.965

Reconnaissance

0.729

0.749

0.696

0.624

0.696

0.349

0.0

F1-Macro

0.926

0.930

0.915

0.776

0.915

0.814

0.779

Accuracy

0.988

0.988

0.988

0.866

0.988

0.977

0.988

Back to article page