Table 8 Performance comparison of different balancing techniques across datasets.

From: Optimizing IoT intrusion detection with cosine similarity based dataset balancing and hybrid deep learning

Datasets

Balancing technique

Accuracy

Precision

Sensitivity

F1 Score

MCC

Markedness

FMI

Time

IoTID20

No balancing

0.9643

0.8329

0.8480

0.8404

0.8204

0.8140

0.8404

403

SMOTE

0.9557

0.9551

0.9565

0.9558

0.9114

0.9114

0.9558

1191

RUS

0.9283

0.9047

0.9575

0.9304

0.8581

0.8596

0.9307

145

Proposed

0.9110

0.9487

0.8712

0.9083

0.8250

0.8271

0.9091

281

N-BaIoT

No balancing

1

1

1

1

1

1

1

133

SMOTE

1

1

1

1

1

1

1

229

RUS

1

0.9999

1

1

0.9999

0.9999

1

249

Proposed

1

1

1

1

1

1

1

81

RT-IoT2022

No balancing

0.9926

0.9790

0.9495

0.9640

0.9601

0.9732

0.9641

223

SMOTE

0.9900

0.9938

0.9861

0.9900

0.9800

0.9800

0.9900

487

RUS

0.9256

0.9442

0.9019

0.9226

0.8520

0.8534

0.9228

57

Proposed

0.9836

0.9900

0.9776

0.9838

0.9673

0.9672

0.9838

92

UNSW Bot-IoT

No balancing

0.5456

0.5206

0.9947

0.6835

0.2213

0.4755

0.7197

3688

SMOTE

0.9157

0.9891

0.8407

0.9089

0.8409

0.8504

0.9119

20285

RUS

0.5589

0.5281

0.9940

0.6898

0.2509

0.4867

0.7245

24

Proposed

0.9507

0.9254

0.9790

0.9514

0.9029

0.9037

0.9518

51

Back to article page