Network impact assessment and stability enhancement through Markov chain analysis and load balancing

Abstract This study proposes a methodology for analyzing and improving network stability using discrete Markov chains. By modeling device transitions in network logs as Markov chains, we developed two models: a Network Impact Assessment Model and a Network Stability Optimization Model. The Network Impact Assessment Model quantifies device impact through three metrics: self-impact, inbound impact, and outbound impact. Chapman–Kolmogorov consistency checks indicated that the first-order transition matrix provides a useful aggregate-level structural approximation (mean absolute error $${<}10^{-2}$$ for all days in both synthetic and real-world datasets), although additional diagnostics revealed non-negligible higher-order dependencies in the real-world data. Analysis of 28-day network logs revealed that certain critical infrastructure devices, particularly authentication servers and core network components, demonstrated significantly high impact values, indicating their crucial role in network operations. To validate the generalizability and robustness of the proposed framework, experiments were conducted on synthetic network data (10,000 devices, 28 days) across ten independent random seeds and three qualitatively distinct network topologies. A sensitivity analysis of the impact weight parameters confirmed that the top-ranked device is stable across all tested configurations and that the Spearman correlation structure varies by at most 0.029, indicating robustness to parameter choice. Comparison with simple fixed-budget top- k centrality baselines on both synthetic and real-world data shows that the proposed framework achieves lower objective values, with the advantage most pronounced for peak load reduction (22.7– $$33.6\\%$$ on synthetic data; $$31.2\\%$$ on real-world data relative to the best centrality baseline; a budget-neutral comparison further confirmed that this advantage holds independently of differences in replication budget). To enhance network stability, we implemented a Network Stability Optimization Model that strategically replicates high-impact devices, optimizing three objective functions independently via simulated annealing. Application to the real-world dataset achieved improvements of $$4.7\\%$$ in device stability, $$7.6\\%$$ in load balancing, and $$56.6\\%$$ in peak load reduction relative to the pre-optimization baseline. The proposed methodology effectively identifies critical network devices and provides a practical load balancing approach through strategic device replication. Our findings contribute to the development of more robust network architectures and offer valuable insights for network administrators managing complex systems.