Table 12 Comparative analysis of the proposed criminal suspect identification model against ML and DL models under adversarial conditions using FGSM on the LFW 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

63.032 ± 0.092

68.867 ± 0.082

69.488 ± 0.031

0.612 ± 0.075

CNBA29

64.549 ± 0.009

70.190 ± 0.041

70.822 ± 0.054

0.620 ± 0.056

FAHP33

65.398 ± 0.062

70.963 ± 0.024

71.587 ± 0.035

0.631 ± 0.085

DL-ACO36

67.511 ± 0.013

72.202 ± 0.068

72.929 ± 0.012

0.692 ± 0.039

DCNN37

73.874 ± 0.056

77.236 ± 0.029

77.920 ± 0.066

0.677 ± 0.051

MTCNN38

76.241 ± 0.064

79.706 ± 0.072

80.399 ± 0.080

0.696 ± 0.062

DNVPT4

81.897 ± 0.075

86.259 ± 0.013

86.773 ± 0.096

0.755 ± 0.038

FECNN39

77.009 ± 0.072

81.085 ± 0.019

81.712 ± 0.054

0.704 ± 0.000

Wavelet2

80.455 ± 0.030

84.602 ± 0.099

85.122 ± 0.035

0.743 ± 0.018

CLSTM40

77.662 ± 0.095

82.087 ± 0.025

82.750 ± 0.008

0.713 ± 0.033

QN-FR41

76.511 ± 0.013

77.202 ± 0.068

78.929 ± 0.012

0.741 ± 0.039

DNN1

79.465 ± 0.020

83.407 ± 0.076

83.991 ± 0.008

0.727 ± 0.012

YOLOv8-FI42

78.144 ± 0.039

79.847 ± 0.094

80.674 ± 0.003

0.742 ± 0.012

FVG-FR43

78.693 ± 0.016

78.923 ± 0.022

79.167 ± 0.053

0.752 ± 0.030

QWE-DNN44

79.408 ± 0.035

80.821 ± 0.041

85.981 ± 0.025

0.767 ± 0.081

FacialCueNet45

74.811 ± 0.094

78.216 ± 0.042

78.933 ± 0.095

0.690 ± 0.009

GAN-DSAEAN 47

80.560 ± 0.070

81.478 ± 0.061

82.324 ± 0.047

0.789 ± 0.076

Proposed

85.380 ± 0.027

89.153 ± 0.064

89.690 ± 0.078

0.792 ± 0.049

Back to article page