Deep Reinforcement Learning for Mitigating Selective Forwarding Attacks in ACO

Wireless Sensor Networks (WSNs) and Internet of Things (IoT) networks rely on resourceconstrained devices and unstable wireless links, making them vulnerable to selective forwarding attacks. Conventional routing methods often prioritise delay, energy and link quality, so a route may remain attractive even when a compromised node drops some packets while forwarding others. This study designed and evaluated a DQN-supported, evidence-gated security mechanism for mitigating selective forwarding attacks in Ant Colony Optimisation (ACO) routing, implemented as the Adaptive Bioinspired Routing Framework (ABRF). The DQN was trained offline on synthetic node-state transitions and, during simulation, used live routing observations to propose one of five graded actions. The evidence gate validated each proposed action using forwarding and route-advertisement evidence before the action modified pheromone or route preference. Evaluation was conducted in a Python discrete-event simulator using smart-home, smart-traffic and wearable-IoT scenarios. The main experiment included 2,250 attack runs and 450 clean-control runs and compared ABRF with simplified RPL-like, AODV-like, ACO-only and traditional-IDS baselines. ABRF achieved a mean attack-step detection rate of 0.847, a hard false-action rate of 0.033 and a packet delivery ratio of 0.665, compared with 0.681, 0.126 and 0.579 for the traditional IDS. Its soft false-action rate was 0.312, indicating that cautious pheromone reductions occurred more often than hard isolation. Dynamic INT8 quantisation reduced the model file size by about 60 percent, although inference was tested on a Colab CPU rather than physical IoT hardware.The findings show that validated DRL security actions can mitigate selective forwarding by directly influencing ACO pheromone decisions while supporting secure route selection under the tested simulation conditions. The conclusions remain limited by the purpose-built simulator, simplified baselines and offline-trained policy.

Authors

Institutions

Publication Details

Journal
International Journal of Innovative Science and Research Technology (IJISRT)
Published
2026-11-16
DOI
https://doi.org/10.38124/ijisrt/26oct1627
Primary Topic
Security in Wireless Sensor Networks
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Deep Reinforcement Learning for Mitigating Selective Forwarding Attacks in ACO

Gaddafi Abdul-Salaam, Ibrahim Samson Salifu
International Journal of Innovative Science and Research Technology (IJISRT)
Security in Wireless Sensor Networks
article

Deep Reinforcement Learning for Mitigating Selective Forwarding Attacks in ACO

Gaddafi Abdul-Salaam, Ibrahim Samson Salifu
article en

Abstract

Wireless Sensor Networks (WSNs) and Internet of Things (IoT) networks rely on resourceconstrained devices and unstable wireless links, making them vulnerable to selective forwarding attacks. Conventional routing methods often prioritise delay, energy and link quality, so a route may remain attractive even when a compromised node drops some packets while forwarding others. This study designed and evaluated a DQN-supported, evidence-gated security mechanism for mitigating selective forwarding attacks in Ant Colony Optimisation (ACO) routing, implemented as the Adaptive Bioinspired Routing Framework (ABRF). The DQN was trained offline on synthetic node-state transitions and, during simulation, used live routing observations to propose one of five graded actions. The evidence gate validated each proposed action using forwarding and route-advertisement evidence before the action modified pheromone or route preference. Evaluation was conducted in a Python discrete-event simulator using smart-home, smart-traffic and wearable-IoT scenarios. The main experiment included 2,250 attack runs and 450 clean-control runs and compared ABRF with simplified RPL-like, AODV-like, ACO-only and traditional-IDS baselines. ABRF achieved a mean attack-step detection rate of 0.847, a hard false-action rate of 0.033 and a packet delivery ratio of 0.665, compared with 0.681, 0.126 and 0.579 for the traditional IDS. Its soft false-action rate was 0.312, indicating that cautious pheromone reductions occurred more often than hard isolation. Dynamic INT8 quantisation reduced the model file size by about 60 percent, although inference was tested on a Colab CPU rather than physical IoT hardware.The findings show that validated DRL security actions can mitigate selective forwarding by directly influencing ACO pheromone decisions while supporting secure route selection under the tested simulation conditions. The conclusions remain limited by the purpose-built simulator, simplified baselines and offline-trained policy.

International Journal of Innovative Science and Research Technology (IJISRT)Vol. 11(10)
Kwame Nkrumah University of Science and Technology (GH), Kiriri Women's University of Science and Technology (KE)
Security in Wireless Sensor Networks
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon