An Intelligent Neural Network Framework for Multi-Company Last-Mile Delivery: Joint Time–Cost Optimization with LSTM Forecasting and Reinforcement Learning Routing
Abstract
In the context of escalating e-commerce demands, last-mile delivery in multi-company logistics environments faces challenges from urban congestion, fragmented operations, and real-time uncertainties, often accounting for over 50% of total supply chain costs. This paper presents a comprehensive neural network framework that integrates Long Short-Term Memory networks for accurate estimated time of arrival forecasting with a reinforcement learning (RL)-based routing engine to achieve joint optimization of delivery time and operational costs across collaborating firms. The modular architecture incorporates a collaboration-aware platform layer that facilitates shared fleet utilization and overlapping zone coordination, processing inputs from logistics databases, real-time traffic APIs, and weather services to enable adaptive decision-making. Theoretical contributions include a Markov Decision Process formulation for multi-agent routing under stochastic constraints, LSTM gating mechanisms enhanced for non-stationary temporal sequences, and a reward function that balances time penalties, fuel costs, and collaboration incentives via weighted multi-objective optimization. Simulations on the Lagos delivery datasets demonstrate 12-15% reductions in joint time-cost metrics compared to baseline heuristics, with robustness to peak-hour volatility. The framework addresses key gaps in existing single-firm, static models by enabling scalable, privacy-preserving multi-stakeholder integration, paving the way for sustainable intelligent transportation systems in dense cities.
Keywords
References
More Articles from INTERNATIONAL JOURNAL OF COMPUTER SCIENCE AND MATHEMATICAL THEORY
Author: Gabriel Odeh Ankeli, PhD, Obekpa Daniel Akor
Author: Ue-Bee, Kakuubari Promise, Sako, D.J.S, Nwiabu, N. D
Author: Collins O. Odondiri, V. I. E. Anireh, O. E. Taylor