TY  - EJOU
AU  - Hassan, Taimoor 
AU  - Hussain, Ibrar 
AU  - Haque, Hafiz Mahfooz Ul 
AU  - Mirza, Hamid Turab 
AU  - Ali, Muhammad Nadeem 
AU  - Kim, Byung-Seo 
AU  - Oh, Changheun 

TI  - Multi-Agent Reinforcement Learning Based Context-Aware Heterogeneous Decision Support System
T2  - Computers, Materials \& Continua

PY  - 2026
VL  - 87
IS  - 3
SN  - 1546-2226

AB  - The expeditious proliferation of the smart computing paradigm has a remarkable upsurge towards Artificial Intelligence (AI) assistive reasoning with the incorporation of context-awareness. Context-awareness plays a significant role in fulfilling users’ needs whenever and wherever needed. Context-aware systems acquire contextual information from sensors/embedded sensors using smart gadgets and/or systems, perform reasoning using reinforcement learning (RL) or other reasoning techniques, and then adapt behavior. The core intention of using an RL-based reasoning strategy is to train agents to take the right actions at the right time and in the right place. Generally, agents are rewarded for the correct actions and punished for incorrect actions. In an RL deployment setting, agents intend to get cumulative maximal rewards through the continuous learning process. These systems often operate in a highly decentralized environment and exhibit complex adaptive behavior. However, the agent’s actions on the imperfect nature of context may cause inconsistent reasoning behavior in terms of the agent’s reward policies. In this paper, we present a semantic knowledge-based Multi-agent Reinforcement Learning (MARL) formalism for a context-aware heterogeneous decision support system. This is a four-layered architecture to schedule user’s routine tasks where user’s data is acquired with limited or no human intervention and perform operations autonomously based on agent’s reward/punishment policies. For this, we develop a comprehensive case study considering three different domains’ ontologies; namely, Smart Home, Smart Shopping, and Smart Fridge Systems, with the prototypal implementation of the system and show the valid execution dynamics, correctness behavior, and verify the agent’s optimal reward policies.
KW  - Multi-agent reinforcement learning; Markov decision process; heterogeneous systems; ontologies; reinforcement learning; context-awareness

DO  - 10.32604/cmc.2026.077510