Neighbor-embedded Graph Neural Network-based Crowd Delivery Traffic Management in Smart City

Kishu Gupta; Deepika Saxena; Ashutosh Kumar Singh; Chung-Nan Lee

cs.CRAug 23, 2026

Neighbor-embedded Graph Neural Network-based Crowd Delivery Traffic Management in Smart City

Authors: Kishu Gupta, Deepika Saxena, Ashutosh Kumar Singh, Chung-Nan Lee

Organizations: Department of Computer Science and Engineering, National Sun Yat-sen University, Kaohsiung, 80424, Taiwan · Department of Computer Science, VIZJA University, Warsaw, 01-043, Poland · Department of Computer Science and Engineering, University of Aizu, Aizuwakamatsu, 9650006, Fukushima Prefecture, Japan · Department of Computer Science and Engineering, Indian Institute of Information Technology Bhopal, 462003, India

Abstract

The significant upsurge in vehicle traffic presents a considerable challenge in the pursuit of smart mobilization and transportation (SMT) worldwide. Current approaches primarily focus on vehicular traffic management through congestion prediction but fall short in addressing essential objectives such as traffic reduction and appropriate vehicle selection to alleviate congestion in smart cities ( $SmCt$ ). To address these concerns, this work introduces a novel Neighbor-Embedded Graph Neural Network-based Crowd Delivery Traffic Management (NeCDM) Model, comprising two key components: the Traffic Congestion Prediction Unit (TCPu) and the Traffic Observation and Management Unit (TOMu). The TCPu utilizes Graph Neural Network (GNN) optimization to accurately predict traffic flow levels at various delivery stations within $SmCt$ ecosystems. Additionally, the TOMu facilitates the intelligent selection of the most suitable delivery vehicles for fulfilling crowd delivery requests ( $CDR$ ). This work emphasizes the potential of crowd delivery as a feasible solution for achieving SMT goals while adhering to smart city parameters ( $\mathcal{SCP}$ s), such as reduced carbon emissions, shorter travel times, and minimized travel distances. The proposed model achieves notable improvements in computational efficiency, including reductions of up to 4.03% in L1 loss ( $£$ ), 16.66% in L2 loss ( $£_{rmse}$ ), and 7.64% in computation time.

Figures & tables

Fig. 1 : Bird eye view of the proposed NeCDM model.

Symbol	Notation	Symbol	Notation
$\mathbb{U}$	User	$DVeh$	Delivery vehicle
$D^{St}$	Delivery station	$SmCt$	Smart city
$CDR$	Crowd delivery request	$\mathfrak{D}$	Distance
$\mathcal{SCP}$	Smart city parameters	$t_{i}$	Time instance
$Ce$	Carbon emission	$TrT$	Travel time
$CTr$	Cost to travel	$S^{St}$	Source station

TABLE I : List of Symbols and Notations

Approach	Method	Focus	Dataset	Contribution	Limitations
Ali et al. [ 3 ] , 2025	Attention-based edge–cloud predictor	Energy-efficient resource allocation; urban traffic prediction	IoV traffic datasets	Attention-based predictor with energy-aware scheduling	Simulation-level validation only; lacks multi-region generalization
Ali et al. [ 1 ] , 2025	Attention-driven spatio-temporal deep hybrid (GCN + temporal)	Traffic flow prediction in transportation CPS	METR-LA, PeMS	Combines spatial attention with temporal encoding for high accuracy	Focuses on accuracy; limited deployment and latency analysis
Ali et al. [ 2 ] , 2024	Resource-aware Multi-Graph Neural Network (MGNN)	Multi-access edge computing; spatio-temporal correlations	Urban traffic datasets	Balances prediction accuracy and edge resource usage	Does not consider service migration or mobility patterns
Ali et al. [ 4 ] , 2021	Attention-based dynamic spatio-temporal network	Dynamic spatial correlations; attention mechanism	Taxi trajectory data	Captures dynamic dependencies between urban regions	Static graph structure limits adaptability
Ali et al. [ 5 ] , 2022	Dynamic Spatio-Temporal GCN	Citywide traffic flow prediction	PeMSD7, TaxiNYC	Models temporal trends with dynamic adjacency	High computational cost
Awan et al. [ 6 ] , 2021	CNN–LSTM hybrid	Urban flow prediction; dynamic correlations	City traffic data	Combines CNN + LSTM for spatio-temporal modeling	Lacks attention or adaptive spatial weighting

TABLE II : Pandect Summary: Related Work on Traffic Flow Prediction and Intelligent Transportation

Fig. 2 : NeCDM model architecture: comprising the TCPu for congestion forecasting across delivery stations, and the TOMu for real-time vehicle selection to optimize crowd delivery, reduce on-road vehicles, and improve smart city parameters.

Pre-processing Parameters		Model Training Parameters
Parameter	Value	Parameter	Value
Minimum observation	1000	Batch size	128
Train set size	80%	Learning rate	0.001
Dropout rate	0.03	weight decay	1 $\times 10^{4}$
Test set size	10%	Epochs	100
Validation set size	10%	Value per epochs	4

TABLE III : Hyper-parameters configuration

Fig. 3 : Training and validation L1-loss performance of the proposed NeCDM model across varying iterations.

Epoch	Loss ( $\pounds$ )		Early	Execution Time
Epoch	$\pounds_{\acute{\Gamma}}$	$\pounds_{\breve{\Gamma}}$	stop	Overall (sec)	It/s
1/100	280.9755	287.1836	00/20	02:00	3.19
10/100	60.7781	53.7435	00/20	01:59	3.21
11/100	56.2907	52.6622	00/20	01:59	3.22
20/100	41.6476	47.3816	01/20	02:00	3.19
21/100	42.0842	47.0381	03/20	01:59	3.21

TABLE IV : Traffic prediction loss and execution time performance optimization for a

CDR

Panel of Figure 8 — Fig. 4 : Traffic flow ( $\varrho$ ) prediction performance across delivery stations over varying timeline.

Panel of Figure 9 — Fig. 5 : Extended route map depicitng Delivery Station ( $D^{St}$ ) travel route.

$\acute{\Gamma}:\breve{\Gamma}:\ddot{\Gamma}$	$Nei_{Ex}$		$Nei_{Em}$
$\acute{\Gamma}:\breve{\Gamma}:\ddot{\Gamma}$	£	$\textbf{\pounds}_{rmse}$	£	$\textbf{\pounds}_{rmse}$
80:10:10	32.71	43.22	29.52	36.32
70:15:15	32.87	43.96	29.64	37.21
50:25:25	33.53	44.23	29.70	37.35

TABLE V : Model efficiency analysis with varying traffic data depicting key performance metrics.

Panel of Figure 11 — Fig. 6 : Traffic prediction model loss analysis for a $CDR$ over varying iteration depicting edge features importance.

Panel of Figure 12 — Fig. 7 : Traffic flow ( $\varrho$ ) prediction performance across $D^{St}$ over varying timeline considering edge features importance.

Approach	Method	Dataset	Loss		Time	It/s	Complexity
Approach	Method	Dataset	£	$\textit{\pounds}_{rmse}$	(sec)	It/s	Complexity
PCNN [ 10 ]	Deep convolution network	Vehicle Passage Records	30.71	45.75	53.93	-	$\mathcal{O}(\Delta tdCN(\eta^{2}+E)(n^{n}))$
LSTM [ 15 ]	Long short term memory	Traffic data from Caltrans PeMS in Oakland	46.73	60.41	58.36	-	$\mathcal{O}(\Delta ntd(2h)(\eta^{2}+E))$
GNN [ 14 ]	Graph with node and edge features	Traffic data form Beijing and Hangzhou city	32.71	43.22	52.72	1.63	$\mathcal{O}(n^{2}NL(\eta^{2}+E)\mathbb{\vec{R}}^{d}\mathbb{\vec{R}}^{d^{\prime}})$
GNN [ 14 ]	Graph without edge features	Traffic data form Beijing and Hangzhou city	31.05	42.37	49.55	1.79	$\mathcal{O}(n^{2}NL(\eta^{2}+E)\mathbb{\vec{R}}^{d})$
NeCDM	Neighbor embedded graph	T-Drive trajectory dataset of Beijing city [ 25 ]	29.52	36.32	46.03	3.21	$\mathcal{O}(nkNL(\eta^{2}+E)\mathbb{\vec{R}}^{d}\mathbb{\vec{R}}^{d^{\prime}})$

TABLE VI : Pandect Comparison: NeCDM model v/s state-of-the-art approaches

Approach	Spatial Modeling	Temporal Modeling	Training Loss	Testing Loss	Time / Epoch (s)	Edge/Cloud Awareness	Predictive Accuracy	Computational Complexity
[ 3 ]	Region-level attention; edge nodes	LSTM / temporal fusion	Medium	Medium	Medium	Yes	Medium-High	$O(N^{2}+NTd)$ (attention + LSTM)
[ 1 ]	Road-network GCN	LSTM / temporal encoding	Low	Low	High	No	High	$O(EF+NTd)$ (GCN + LSTM)
[ 2 ]	Multi-graph convolution	Temporal CNN / GRU	Low	Low	High	Yes	High	$O(M\cdot EF+NTd)$ (multi-graph + temporal)
[ 4 ]	Dynamic adjacency matrices	LSTM	Low	Low	Medium	No	High	$O(N^{2}+NTd)$ (attention + LSTM)
[ 5 ]	GCN layers + adjacency update	LSTM / temporal CNN	Low	Low	High	No	High	$O(EF+NTd)$ (dynamic GCN + LSTM)
[ 6 ]	Grid-based CNN	LSTM	Medium	Medium	Medium	No	Medium	$O(W^{2}D+NTd)$ (CNN + LSTM)

TABLE VII : Comprehensive Comparison: NeCDM v/s SoTA approaches depicting model metrics and computational features

Intelligent Transportation Systems

Intelligent Transportation Systems apply sensing, communication, AI, and optimization to make transport networks safer and more efficient.

See all 301 papers

GNN-Guided Network Reduction for Centralized Urban Traffic Routing
Oct 27, 2026Nikolai Antonov, Lukáš Chrpa, Andrii Nyporko +1Vehicle Routing Intelligent Transportation Systems
Czech Technical University in Prague, Czech Institute of Informatics, Robotics and Cybernetics, Jugoslávských partyzánů 1580/3, Prague, 160 00, Czech Republic · Czech Technical University in Prague, Faculty of Electrical Engineering, Technická 2, Prague, 166 27, Czech Republic
ButterMamba: Butterworth-Enhanced Spatial-Temporal Mamba for Efficient Traffic Flow Prediction
Aug 30, 2026Limiao Zhang, Yuhui Lu, Jie Gao +3Mamba Intelligent Transportation Systems
State Key Laboratory of Opto-Electronic Information Acquisition and Protection Technology, Anhui University · Institute of Physical Science and Information Technology, Anhui University · School of Innovation and Entrepreneurship, Shandong University +1
Parameter Efficient Hybrid Transformer (PEHT) for Network Traffic Prediction via Dynamic Urban Congestion Integration
Jun 26, 2026Abdolazim Rezaei, Mehdi Sookhak, Mahboobeh HaghparastMultivariate Time Series Forecasting Traffic Forecasting
Department of Computer Science Texas A&M University Corpus Christi, USA

Oct 27, 2026J/K move · Enter open · S save

GNN-Guided Network Reduction for Centralized Urban Traffic Routing

Nikolai Antonov, Lukáš Chrpa, Andrii Nyporko, Jan Hůla

Czech Technical University in Prague, Czech Institute of Informatics, Robotics and Cybernetics, Jugoslávských partyzánů 1580/3, Prague, 160 00, Czech Republic · Czech Technical University in Prague, Faculty of Electrical Engineering, Technická 2, Prague, 166 27, Czech Republic

Vehicle Routing Intelligent Transportation Systems Graph Sparsification Graph Neural Networks

Open paper AskPDF arXiv

We consider the centralized traffic routing (CTR) problem, where a central planner assigns routes to origin-destination (OD) vehicle flows to minimize congestion-dependent segment cost. Prior work formulates CTR efficiently as a mixed-integer linear program (MILP), but scaling this formulation to the full road network is computationally prohibitive within an operational time budget. Bounded Routes and Smart Routes speed up optimization using precomputed reduced networks; however, these methods do not account for vehicle information within each OD flow. Motivated by this gap, we propose a learning-based urban network reduction approach. A dedicated graph neural network (GNN) first scores road segments for each OD flow using the road network and vehicle-flow information. Highly ranked segments are then incorporated into complete OD routes, and the original segment-flow MILP is solved on the resulting reduced road network. The GNN thus guides the reduction of the road network, while the solver enforces feasibility and determines the routing. Experiments on four real urban networks -- Prague-Dejvice, London, New York, and Sydney -- show that the proposed approach achieves the smallest relative objective difference to the

30

-second full-graph reference in every city under both traffic regimes. Averaged across cities, this difference is more than six times lower than that of the best competing method. Moreover, traffic simulation results show that the optimization gains of the proposed method lead to more efficient routing, with the highest aggregate arrival rate and a competitive trip duration per arrival percentage point.

Abstract

Intelligent Transportation Systems

Explore similar work

GNN-Guided Network Reduction for Centralized Urban Traffic Routing

ButterMamba: Butterworth-Enhanced Spatial-Temporal Mamba for Efficient Traffic Flow Prediction

Parameter Efficient Hybrid Transformer (PEHT) for Network Traffic Prediction via Dynamic Urban Congestion Integration