Analysis of Encrypted Network Traffic for Enhancing Cyber-security in Dynamic Environments

Figures & data

Figure 1. Overall architecture of the proposed method.

Figure 2. Tree-based Spider-Net multipath.

Figure 3. Network traffic identification process based on DPI.

Figure 4. Deep packet inspection with AES.

Figure 5a. Deep reinforcement learning.

Table 1. Tabu search algorithm parameters.

Parameters	Values
Tabu List Size	7
Number of Neighbor	6
Aspiration Level	0.05
Number of Iterations	100

Table

Algorithm 1 TS algorithm utilizing a context of packet services in reward maximization.

1. Initialization: The initial solution can be mentioned as $B_{\mathit{init}}$. 2. $\mathit{B}\leftarrow B_{\mathit{init}}$ 3. Let g(B) be the reward achieved using a result outcome. $g_{\mathit{max}}\leftarrow \mathit{g}\left(B_{\mathit{init}}\right)$ 4. while Number of Neighbourhood iterations $\le$ maxiter do 5. Formation of Neighbourhood: Every possible neighbor of TS in initial solution outcome B is created, expected can be listed as tabu. 6. Selection of Neighbor: The chosen maximum reward solution achieves B′ based on the neighbors set, and the condition $\mathit{B}{ }^{\mathrm{\prime }}$ is not listed as Tabu, $\mathit{B}\leftarrow \mathit{B}{ }^{\mathrm{\prime }}$. 7. Updation of Tabu list: The reward $\mathit{g}\left(\mathit{B}{ }^{\mathrm{\prime }}\right)$ is added corresponding to the fixed solution $\mathit{B}{ }^{\mathrm{\prime }}$ is added to the TL. 8. Updation of Maximum Reward: The maximum obtained of a $g_{\mathit{max}}$reward can be updated as follows, if $\mathit{g}\left(\mathit{B}{ }^{\mathrm{\prime }}\right)>g_{\mathit{max}}$, then $g_{\mathit{max}}\leftarrow \mathit{g}\left(\mathit{B}{ }^{\mathrm{\prime }}\right)$ end if 9. end while loop

Table

Algorithm 2 Proximal Policy Optimization (PPO)

1. Initialize an actor as $\mathit{\mu }$: $\mathit{D}\to F^{\mathit{m}+1}$ and $\mathit{\sigma }:\mathit{D}\to \mathit{diag}\left({\sigma }_1,{\sigma }_2,\dots ,{\sigma }_{\mathit{m}+1}\right)$ 2. for i=1 to M do Run on a policy ${\pi }_{\theta }\sim \mathit{N}\left(\mathit{\mu }\left(\mathit{D}\right),\mathit{\sigma }\left(\mathit{D}\right)\right)$ for time steps T and collect a $\left(D_t,\mathit{ht},r_t\right)$ Advantages estimate a $\widehat{A_t}={\sum }_{\mathit{t}{\mathit{ }}^{\mathrm{\prime }}>\mathit{t}}{\gamma }^{\mathit{t}{\mathit{ }}^{\mathrm{\prime }}-\mathit{t}}{r}_{\mathit{t}{\mathit{ }}^{\mathrm{\prime }}}-\mathit{v}\left(D_t\right)$ Old policy update ${\pi }_{\mathit{old}}\leftarrow {\pi }_0$ 3. for j=1 to N do Actor policy updation use of policy gradient: ${\sum }_i{\mathrm{\nabla }}_{\theta }{L}_i^{\mathit{clip}}\left(\mathit{\theta }\right)$ Critic updation is: $\mathrm{\nabla L}\left(\mathrm{\varnothing }\right)=-{\sum }_{t=1}^T\mathrm{\nabla }{\widehat{A_t}}^2$ 4. end for 5. end for

Figure 5b. Standard convolution.

Figure 5c. Depth-wise convolution.

Figure 6. Lightweight unit B architecture.

Figure 7. Channel shuffle.

Figure 8. Fitness value of HGS-ROA.

Figure 9. Number of users vs authentication rate (bytes).

Table 2. Numerical outcomes of authentication rate.

Number of users (x-axis)	Authentication rate (bytes)-(y-axis)
	FML	HIDT	Proposed
10	130	155	180
20	160	180	210
30	190	220	240
40	220	235	270
50	230	250	300

Figure 10. Number of IOT devices vs accuracy (%).

Table 3. Numerical outcomes of accuracy.

Number of IOT devices (x-axis)	Accuracy(%) - (y-axis)
	FML	HIDT	Proposed
10	60	67	76
20	70	75	83
30	78	85	90
40	82	89	94
50	90	95	99

Figure 11. Number of IOT devices vs precision (%).

Table 4. Numerical outcomes of precision.

Number of IOT devices (x-axis)	Precision(%) - (y-axis)
	FML	HIDT	Proposed
10	60	68	75
20	70	75	82
30	78	80	87
40	84	89	94
50	89	94	98

Figure 12. Number of IOT devices vs throughput (kbps).

Table 5. Numerical outcomes of throughput.

Number of IOT devices (x-axis)	Throughput(kbps) - (y-axis)
	EHGSO	HIDT	Proposed
10	1.5	2.9	3.5
20	2.9	3.8	4.5
30	3.5	4	5.8
40	3.8	4.5	6
50	4	5.2	7

Figure 13. Number of IOT devices vs packet delivery ratio (%).

Table 6. Numerical outcomes of packet delivery ratio.

Number of IOT devices (x-axis)	PDR(%) - (y-axis)
	EHGSO	HIDT	Proposed
10	57	63	70
20	63	69	76
30	70	77	82
40	79	84	87
50	85	88	92

Figure 14. Number of IOT devices vs attack detection rate (%).

Table 7. Numerical outcomes of attack detection rate.

Number of IOT devices (x-axis)	Attack Detection Rate(%) - (y-axis)
	FML	HIDT	Proposed
10	60	65	74
20	72	76	81
30	78	83	89
40	83	87	93
50	91	95	98

Figure 15. Number of IOT devices vs delay (sec).

Table 8. Numerical outcomes of delay.

Number of IOT devices (x-axis)	Delay (sec) - (y-axis)
	FML	HIDT	Proposed
10	4.5	3.9	3
20	3.4	2.7	2.2
30	2.4	2	1.6
40	1.6	1.2	0.8
50	0.9	0.6	0.3

Figure 16. Number of IOT devices vs recall (%).

Table 9. Numerical outcomes of recall.

Number of IOT devices (x-axis)	Recall (%) - (y-axis)
	FML	HIDT	Proposed
10	58	68	75
20	68	74	80
30	76	80	86
40	82	87	93
50	88	92	97

Figure 17. Number of IOT devices vs F1-score (%).

Table 10. Numerical outcomes of F1-score.

Number of IOT devices (x-axis)	Recall (sec) - (y-axis)
	FML	HIDT	Proposed
10	60	69	74
20	69	75	80
30	71	78	85
40	79	85	91
50	85	90	96