Open Access

ARTICLE

Ensemble-Guided Pseudorandom Number Generation with Residue Number System Transformation

Issah Zabsonre Alhassan^1,2,*, Gaddafi Abdul-Salaam¹, Michael Asante¹, Yaw Marfo Missah¹, Alimatu Sadia Shirazu¹

1 Department of Computer Science, Faculty of Physical and Computational Sciences, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana
2 Department of Medical Imaging, School of Allied Health Sciences, University for Development Studies, Tamale, Ghana

* Corresponding Author: Issah Zabsonre Alhassan. Email:

Journal of Cyber Security 2026, 8, 577-607. https://doi.org/10.32604/jcs.2026.085303

Received 08 May 2026; Accepted 23 July 2026; Issue published 21 August 2026

Abstract

A hybrid approach to pseudorandom number generation that couples ensemble learning with the Residue Number System (RNS) is presented in this paper. Unlike conventional deterministic generators that depend solely on direct algorithmic transformation, the proposed method first maps a seed-driven integer sequence into its RNS representation under the coprime moduli set {3, 5, 7, 11}, whose dynamic range is M = 1155, thereby introducing modular non-linearity through a static, stateless feature transformation. A soft-voting ensemble of Logistic Regression, Random Forest, and Support Vector Machine then serves as a decision layer that classifies and re-maps the transformed values into the output sequence. To improve rigour and reproducibility, this revised version reports the full experimental workflow, the hyperparameter-tuning procedure, an ablation study, a quantitative and qualitative comparison with previously reported Pseudorandom Number Generators (PRNGs), an explicit separation of one-time training cost from online generation cost, and an extended security-oriented assessment. The generated sequences were evaluated using classification metrics, Shannon entropy, histogram and autocorrelation analyses, the Kolmogorov-Smirnov test, and an expanded NIST SP 800-22 battery applied to one million bits. Across these tests, the generator produced statistically acceptable sequences with high entropy, low serial dependence (Kolmogorov-Smirnov D = 0.0225, p = 0.6819) and agreement with all eight NIST tests examined, while a single-bit input perturbation produced an avalanche response close to 49.6%. The study nonetheless makes clear that strong statistical performance does not by itself imply cryptographic security and that the present Python prototype is not throughput-competitive with optimized cryptographic generators such as ChaCha20 or AES-CTR. The method is therefore positioned as a machine-learning-assisted pseudorandom generation framework and a basis for further cryptographic study, rather than as a fully validated cryptographically secure pseudorandom number generator.

---

Keywords

---