Open Access

ARTICLE

Nonlinear Fractional Computer Virus Propagation in Safety Critical Heterogeneous Networks Analysis with Surrogate Deep Neuroarchitecture

Kiran Asma, Muhammad Asif Zahoor Raja^*

Graduate School of Engineering Science and Technology, National Yunlin University of Science and Technology, Douliu, Taiwan

* Corresponding Author: Muhammad Asif Zahoor Raja. Email:

(This article belongs to the Special Issue: Emerging Technologies in Information Security: Modeling, Algorithms, and Applications)

Computer Modeling in Engineering & Sciences 2026, 148(1), 46 https://doi.org/10.32604/cmes.2026.083532

Received 20 April 2026; Accepted 27 May 2026; Issue published 27 July 2026

Abstract

The accelerated digital transformation of critical infrastructure has yielded unprecedented system interconnectivity, enhancing operational efficiency, simultaneously expanding the epidemiological propagation surface in heterogeneous networks. A novel machine learning-driven neuroarchitecture is designed in the present study, leveraging multilayer autoregressive exogenous neural networks (ARXNNs) iteratively trained with the Levenberg Marquardt (LM) algorithm, i.e., ARXNNs-LM, to address the intricate temporal dynamics of nonlinear fractional epidemiological computer virus propagation in the networks. The proposed ARXNNs-LM methodology effectively models the dynamic state transitions between susceptible, infected, and recovered systems. The dataset is synthesized through the application of the Grünwald–Letnikov (GL) fractional finite difference to numerically solve the nonlinear fractional computer virus propagation (FCVP) model for sundry case studies, such as variation in the infection rate of susceptible computers, the recovery rate of infected computers, and the removal rate from the network, while maintaining the fixed external computers connection rate. The simulated information is arbitrarily divided into training, testing, and validation subsets, while the network optimization is achieved by minimization of mean squared error (MSE), with achieved error magnitude around 10⁻⁹ to 10⁻¹³. The proposed ARXNNs-LM methodology is compared with referenced numerical solutions of the FCVP model in terms of MSE convergence, absolute deviation from actual values, error frequency histograms, cross and autocorrelation analysis, optimization control parameters analysis, and time series analysis, to highlight the accuracy, robustness, and resilience. The precision and stability of the designed technique are further validated through the analysis of variance (ANOVA) assessment across multiple optimization algorithms applied to the FCVP model.

---

Keywords

---