Table 11 Comparative analysis of the proposed criminal suspect identification model against ML and DL models under adversarial conditions using FGSM on the CelebA dataset.

From: Reinforcement learning-driven feature selection enhanced by an evolutionary approach tuning for criminal suspect identification

Model

Accuracy

F-measure

G-means

AUC

MISSL28

59.565 ± 0.075

65.169 ± 0.053

66.030 ± 0.034

0.601 ± 0.028

CNBA29

61.306 ± 0.046

66.657 ± 0.031

67.429 ± 0.082

0.610 ± 0.041

FAHP33

62.778 ± 0.066

68.372 ± 0.009

69.128 ± 0.057

0.616 ± 0.096

DL-ACO36

64.511 ± 0.044

69.202 ± 0.023

70.929 ± 0.007

0.641 ± 0.035

DCNN37

71.563 ± 0.018

75.034 ± 0.054

75.562 ± 0.027

0.678 ± 0.022

MTCNN38

73.730 ± 0.086

77.566 ± 0.048

78.055 ± 0.057

0.706 ± 0.014

DNVPT4

79.547 ± 0.093

82.697 ± 0.032

83.116 ± 0.027

0.755 ± 0.084

FECNN39

74.662 ± 0.034

78.370 ± 0.068

78.899 ± 0.030

0.715 ± 0.055

Wavelet2

78.064 ± 0.049

81.270 ± 0.010

81.742 ± 0.091

0.750 ± 0.083

CLSTM40

76.042 ± 0.076

79.815 ± 0.073

80.299 ± 0.029

0.722 ± 0.063

QN-FR41

75.962 ± 0.044

77.406 ± 0.024

78.984 ± 0.008

0.736 ± 0.054

DNN1

77.228 ± 0.009

80.598 ± 0.096

81.065 ± 0.038

0.737 ± 0.054

YOLOv8-FI42

77.504 ± 0.079

78.143 ± 0.100

79.699 ± 0.043

0.746 ± 0.078

FVG-FR43

78.113 ± 0.058

79.928 ± 0.006

80.518 ± 0.077

0.749 ± 0.063

QWE-DNN44

79.539 ± 0.027

80.007 ± 0.054

81.611 ± 0.036

0.763 ± 0.057

FacialCueNet45

72.460 ± 0.067

75.975 ± 0.081

76.488 ± 0.083

0.690 ± 0.058

GAN-DSAEAN47

80.803 ± 0.082

81.038 ± 0.054

82.635 ± 0.085

0.726 ± 0.086

Proposed

84.227 ± 0.019

86.078 ± 0.092

87.441 ± 0.056

0.765 ± 0.010

Back to article page