Open Access

ARTICLE

Dendritic Cell Algorithm with Reinforcement Learning for Adaptive Signal Categorization

Yousra Abudaqqa^*, Zulaiha Ali Othman, Azuraliza Abu Bakar

Research Center for Artificial Intelligent Technology, Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia, Bangi, Selangor, Malaysia

* Corresponding Author: Yousra Abudaqqa. Email:

Computer Modeling in Engineering & Sciences 2026, 147(2), 37 https://doi.org/10.32604/cmes.2026.079034

Received 13 January 2026; Accepted 26 March 2026; Issue published 27 May 2026

Abstract

Signal categorization is a critical component of the Dendritic Cell Algorithm (DCA), as it directly influences its anomaly detection capability. Conventional DCA implementations typically rely on heuristic or optimization-based approaches, such as Grouping Particle Swarm Optimization (GPSO), Grouping Genetic Algorithms (GGA), Principal Component Analysis (PCA), and Support Vector Machines (SVM), to determine mappings between input features and the three immunological signal categories: Pathogen-Associated Molecular Patterns (PAMP), Danger Signals (DS), and Safe Signals (SS). These approaches depend heavily on domain expertise and predefined rules, making the resulting signal mappings static and often dataset specific. Consequently, the traditional DCA lacks flexibility across diverse data domains and may fail to capture evolving patterns in complex datasets. To address this limitation, this study integrates Reinforcement Learning (RL) into the DCA framework to develop an adaptive signal categorization mechanism. The proposed RL-DCA model employs a Q-learning agent to dynamically assign features to the three signal categories based on reward feedback derived from classification performance. Through continuous interaction with the environment, the RL agent learns an optimal signal mapping policy that improves the quality of generated signals while reducing reliance on manually defined configurations. Experimental evaluations conducted on nine benchmark datasets from multiple domains demonstrate that the proposed RL-DCA framework consistently outperforms existing DCA variants in terms of anomaly detection accuracy and robustness. The results confirm that reinforcement learning provides an effective mechanism for enabling adaptive and data-driven signal categorization in immune-inspired anomaly detection systems.

---

Keywords

---