Markov Decision Process and Double Deep Q-Network-Based Sequential Decision Optimization for Cellphone Return Management in Reverse Logistics
Big Data and Cognitive Computing · 14.11.2026 · 10 (11) · 390
DOI: 10.3390/bdcc10110390 ↗Kurz gefasst
Fachartikel in der Zeitschrift Big Data and Cognitive Computing, online erschienen am 14.11.2026. Verfasst von Abdul Aziz Ibrahim und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Natural Sciences and Engineering Research Council of Canada.
Abstract (Originalsprache)
Cellphone return rates remain low despite increasing awareness of electronic waste and established recycling programs. Existing reverse logistics strategies rely on static rules that fail to adapt to sequential decision-making under uncertainty, resulting in suboptimal return performance. To address this, the problem is formulated as a Markov Decision Process (MDP) and solved using a Double Deep Q-Network (DDQN) that learns an adaptive intervention policy. The agent selects five actions: no action, request feedback, offer incentive, perform data deletion, and refer to service. Evaluated on 500 held-out episodes (10,000 decision steps), the learned policy achieves a customer return success rate of 20.6% (103/500), compared with 8.50% for a rule-based policy and 4.43% for a random policy, representing a 4.6× improvement over random selection. The model reaches a stable cumulative reward of 5.04 ± 0.03 during training. Training analysis shows cumulative reward convergence within approximately 100 episodes; by episode 300, the training convergence metric reached 0.95–1.0, while Q-value standard deviation stabilized at 0.08–0.12 and entropy decreased from 1.6 to 0.3–0.6. Action analysis over 10,000 decisions shows data deletion achieved 15.53% (3.1× baseline), incentives 12.33% (2.5× baseline), and communication-based actions 5.87–6.57%. Cross-seed analysis confirms stable performance across random seeds, with 0.50% variation (5.0462 ± 0.0251). Data security sensitivity analysis improved customer return success from 20.6% to 21.45%, representing a 0.85 percentage-point increase (4.1% relative improvement). The proposed DDQN enables adaptive intervention selection in cellphone reverse logistics.
Förderung
- Natural Sciences and Engineering Research Council of Canada
Zitieren
APA
Ibrahim, A. A., & Abdul-Kader, W. (2026). Markov Decision Process and Double Deep Q-Network-Based Sequential Decision Optimization for Cellphone Return Management in Reverse Logistics. Big Data and Cognitive Computing, 10(11), 390. https://doi.org/10.3390/bdcc10110390
Vancouver
Ibrahim AA, Abdul-Kader W. Markov Decision Process and Double Deep Q-Network-Based Sequential Decision Optimization for Cellphone Return Management in Reverse Logistics. Big Data and Cognitive Computing. 2026;10(11):390. doi:10.3390/bdcc10110390
BibTeX
@article{Ibrahim2026,
author = {Ibrahim, Abdul Aziz and Abdul-Kader, Walid},
title = {Markov Decision Process and Double Deep Q-Network-Based Sequential Decision Optimization for Cellphone Return Management in Reverse Logistics},
journal = {Big Data and Cognitive Computing},
year = {2026},
volume = {10},
number = {11},
pages = {390},
doi = {10.3390/bdcc10110390}
}
EUVIMED