4.2 Comparative analysis
Table 5 shows the computing service rate of the proposed model. Where the existing models such as RF, LQ (Least Quece), KNN, and Q-learning achieved CSR of 0.5, 0.9, and 0.7, 15. However, the proposed model achieved a CSR rate of 20. The graphical representation of CSR is depicted in Fig. 5.
Table 5
Comparative Analysis of CSR
Method | CSR |
Random Forest | 0.5 |
Least Queue | 0.9 |
KNN | 0.7 |
Q-Learning | 15 |
Modified Deep CNN- BiLSTM | 20 |
Table 6. shows the average delay obtained by the proposed model is 2, whereas the existing models, like RF, LQ, KNN, and QL, attained average delays of 15, 14, 12, and 5. The graphical representation of the average delay is illustrated in Fig. 6.
Table 6
Comparative analysis of average delay
Method | Average Delay |
Random Forest | 15 |
Least Queue | 14 |
KNN | 12 |
Q-Learning | 5 |
Modified Deep CNN- BiLSTM | 2 |
Likewise, the model's probability is denoted by using different methods like RF, LQ, KNN, and Q-Learning, and the probability obtained are 0.05, 0.04, 0.03, 0.02, however, the proposed has achieved the lowest value of 0.01 is depicted in Table 7. Graphical illustration of the table is denoted in Fig. 7.
Table 7
Comparative analysis of probability
Method | Probability |
Random Forest | 0.05 |
Least Queue | 0.04 |
KNN | 0.03 |
Q-Learning | 0.02 |
Modified Deep CNN- BiLSTM | 0.01 |
Correspondingly, Table 8 depicts the Handover Failure Rate (HFR) of the proposed model obtained as 1ms. In contrast, the existing methods like DRL-MBP, DRL-SBP, Static, adaptive, QL, and no MLB (Mobility Load Balancing) achieved HFR of 4, 6, 7, 8, 9, and 10. A graphical representation of the table is depicted in Fig. 8.
Table 8
Comparative analysis of handover failure rate
Method | HFR (ms) |
DRL-MBP | 4 |
DRL-SBP | 6 |
Static | 7 |
Adaptive | 8 |
Q-Learning | 9 |
no MLB | 10 |
Modified Deep CNN- BiLSTM | 1 |
The handover failure rate caused by the proposed model is low. At the same time, existing models caused a very high handover rate. The handover Failure Rate is caused by various reasons, such as poor configuration, poor convergence or fluctuating signal strength. Therefore, the lower the HFR, the higher the model's performance.
$$HFR=\frac{{N}_{HO\_fail}}{{N}_{HO\_fail}+{N}_{HO\_succ}}$$
Similarly, the Global Load Distribution (GLD) of the existing studies are tabulated in Table 9. Where the GLD obtained by the current models such as DRL-MBP, DRL-MBP, DRL-SBP, Static, adaptive, QL, and No-MLB achieved GLD of 0.15, 0.12, 0.18, 0.14, 0.22, 0.23 and proposed model of 0.1. Figure 9 shows the graphical representation of GLD.
Table 9
Comparative analysis of GLD
Method | GLD |
DRL-MBP | 0.15 |
DRL-SBP | 0.12 |
Static | 0.18 |
Adaptive | 0.14 |
Q-Learning | 0.22 |
No MLB | 0.23 |
Modified Deep CNN- BiLSTM | 0.1 |
The accuracy of the existing models is compared using different models such as Decision Tree, RF, CNN, FCNN, hybrid Adaboost and RF and is tabulated in Table 10. However, the proposed model delivered an accuracy rate of 99.96, whereas the accuracy obtained by the existing models was low. Graphical representation of the model is illustrated in Fig. 10.
Table 10
Comparative analysis of accuracy
Methods | Accuracy |
Decision Tree | 80% |
Random Forest | 81% |
CNN | 82% |
F-CNN | 92% |
Hybrid Adaboost and RF (Phase 3) | 98% |
Modified Deep CNN- BiLSTM | 99.96% |
Table 11 shows the comparison of different metrics with the proposed model and state-of-art approaches. The metrics considered are energy efficiency, network stability, QoS, and user experience of the proposed and existing models.
Table 11
Comparative analysis of different metrics [69]
Metrics | Existing % | Proposed - Modified Deep CNN- BiLSTM % |
Energy efficiency (Network Lifetime) | 88.03% | 92.3% |
Network stability | 98.47% | 99.8% |
Network scalability (Network Capacity) | 95.76% | 96.78% |
QoS | 93.15% | 99.2% |
User Experience | 100% | 100% |
From Table 11, it can be identified that the proposed model has delivered better outcomes for all metrics when compared to state-of-art approaches by obtaining a network lifetime of 92.3%, network stability of 99.8%, network capacity of 96.78%, QoS of 99.2% and it is due to the incorporation of Modified Deep CNN-BiLSTM in the proposed work. However, user experience attained by both existing and proposed work is 100%. Graphical representation is shown in Fig. 11.
Similarly, the Latency and Network lifetime of the existing and proposed Modified Deep CNN-BiLSTM are shown in Table 12 Latency obtained by the existing model is 6 sec, whereas the proposed model is 3.2 sec, as the lower the latency, the better the performance of the model as lower latency refers to the minimal delay. Likewise, the Network lifetime is also depicted in Table 6.8, where the proposed work has attained a network lifetime of 96.78%, whereas existing work has attained a network lifetime of 88% which shows the reliability of the model.
Table 12
Comparative analysis of latency and network
Method | Existing | Proposed- Modified Deep CNN- BiLSTM |
Latency (seconds) | 6 sec | 3.2 sec |
Network lifetime (mbps) | 88 | 96.78 |
Networks with a better lifetime are less likely to experience incessant disruptions, which eventually influence the accuracy of the model for finding efficient routes and classification of handover and load balancing. Pictorial representation is highlighted in Fig. 12. Corresponding to other metrics, Throughput is also compared between the existing and proposed Modified Deep CNN-BiLSTM model. The existing model has obtained a throughput value of 350000 s whereas the throughput of the proposed model is 380000 s. When the throughput of the network is high, it is detected that a huge volume of data can be transmitted within a particular time frame. Thus, the high throughput value of the proposed model ensures that packets can be sent and received without any delay, thereby enabling swift communication between each node. The illustration representation of Table 13 is shown in Fig. 13.
Table 13
Comparative analysis of throughput [70]
Method | Existing | Proposed- Modified Deep CNN- BiLSTM |
Throughput (Mbps) | 350000 | 380000 |