Open research questions in Traffic Prediction and Management Techniques

What the literature leaves open

• A primary objective of the proposed framework is not only to predict accident severity but also to utilize this information to support safer route selection. To achieve this objective, route generation was formulated as a multi-criteria optimization problem that simultaneously considers travel efficiency and accident risk. Unlike conventional navigation systems that prioritize the shortest or fastest route, the proposed framework incorporates accident-risk information into the route-selection process to reduce user exposure to hazardous road segments. The route-optimization process is guided by a composite objective function that combines travel cost and route-risk information. Two weighting parameters, α and β, are used to control the relative importance of these components. The parameter α represents the contribution of route efficiency, including travel distance and estimated travel time, whereas β represents the contribution of accident risk derived from historical accident records and severity predictions. By adjusting these parameters, the framework can balance the competing objectives of transportation efficiency and road safety. The values of α and β were determined through system development. empirical validation during Page 17 of 26 Multiple candidate parameter combinations were evaluated to examine their influence on route characteristics, including travel distance, estimated travel time, and cumulative route-risk score. Configurations assigning excessive importance to efficiency produced shorter routes but frequently traversed accident-prone locations. Conversely, configurations strongly emphasizing safety generated routes with lower risk but introduced unnecessary increases in travel distance and travel duration. The final parameter configuration was selected because it consistently achieved a favourable compromise between these competing objectives, providing meaningful risk reduction while maintaining practical travel efficiency. To further assess parameter stability, a sensitivity analysis was performed using several candidate weight combinations. The analysis demonstrated that the routing framework maintained consistent behaviour across a reasonable range of parameter settings. Although route characteristics varied slightly as the relative importance of efficiency and safety changed, the overall ranking of safer routes remained stable. These findings indicate that the selected parameter configuration provides reliable performance and is not overly sensitive to minor variations in the weighting factors. To validate the effectiveness of the proposed routing strategy, a quantitative comparison was conducted between a conventional shortest-path routing approach and the proposed risk-aware route recommendation framework. The shortest-path approach selected the route with the minimum travel distance between the origin and destination. In contrast, the proposed framework incorporated accident-severity predictions and spatialrisk information into the route-selection process. Road segments associated with elevated accident risk were assigned higher traversal costs, allowing the routing algorithm to identify alternative paths that reduced overall exposure to hazardous locations. Table 9 presents the comparative evaluation. Three performance criteria were considered: travel distance, estimated travel time, and cumulative route-risk score. The route-risk score was calculated by aggregating the risk values associated with the road segments contained within each candidate route. The results demonstrate that the proposed routing strategy substantially reduces cumulative route risk.

  Data-driven road safety enhancement: neural network–based accident classification and safe route identification using spatial network analysis · 2026 · DOI

• This work presents a horizon-aware hybrid forecasting ap- proach for smart parking systems using sensor-based oc- cupancy data. By systematically evaluating a diverse set of classic machine learning and deep learning models under a unified experimental protocol in two real-world scenarios, the study demonstrated that predictive accuracy varies sub- stantially across forecast horizons and that no single model is universally optimal over a 24-hour range. To address this challenge, we introduced a Global Weighted Blend Hybrid and a Local Weighted Blend Hybrid forecasting strategy. Experimental results on two parking facilities showed that both hybrid variants outperform individual baseline mod- els, with the Local Weighted Blend Hybrid achieving the best overall performance. These results suggest that explic- itly accounting for horizon-dependent model behavior could lead to more accurate and robust multi-step forecasts. Beyond methodological contributions, the proposed ap- proach was developed and evaluated within an end-to-end data engineering pipeline, emphasizing deployment feasi- bility and decision-support relevance. By translating multi- horizon predictions into interpretable next-day availability trajectories, the system could support informed arrival-time planning and enhance the practical usefulness of parking information services. Datenbank-Spektrum To further study the feasibility of this approach, we plan to integrate this approach into an adaptive architecture for real-time inference. This architecture would manage mul- tiple machine learning models and dynamically adapt the data pipelines to the workload of downstream tasks (here: different prediction horizons). Since using context informa- tion could provide higher accuracy than any single-model approach, we propose to structure the system as an adaptive system implementing the MAPE-K loop (Monitor, Analyze, Plan, Execute, over a shared Knowledge base), a reference model for engineering feedback control in autonomic com- puting [36]. Overall, this study highlights the importance of moving beyond single-model forecasting in multi-step prediction settings and demonstrates that horizon-aware hybridization could be a simple yet effective strategy for improving pre- dictive performance in real-world smart city applications. Funding Open Access funding enabled and organized by Projekt DEAL. Data availability All data supporting the findings of this work are in- cluded in the article. Open Access Dieser Artikel wird unter der Creative Commons Na- mensnennung 4.0 International Lizenz veröffentlicht, welche die Nutzung, Vervielfältigung, Bearbeitung, Verbreitung und Wieder- gabe in jeglichem Medium und Format erlaubt, sofern Sie den/die ursprünglichen Autor(en) und die Quelle ordnungsgemäß nennen, einen Link zur Creative Commons Lizenz beifügen und angeben, ob Änderungen vorgenommen wurden. Die in diesem Artikel enthaltenen Bilder und sonstiges Drittmaterial unterliegen ebenfalls der genannten Creative Commons Lizenz, sofern sich aus der Abbildungslegende nichts anderes ergibt. Sofern das betreffende Material nicht unter der genannten Creative Commons Lizenz steht und die betreffende Hand- lung nicht nach gesetzlichen Vorschriften erlaubt ist, ist für die oben aufgeführten Weiterverwendungen des Materials die Einwilligung des jeweiligen Rechteinhabers einzuholen. Weitere Details zur Lizenz ent- nehmen Sie bitte der Lizenzinformation auf http://creativecommons. org/licenses/by/4.0/deed.de.

  Data Engineering for Horizon-Aware Smart Parking Systems · 2026 · DOI

• Abstract Glare from both low sun angles under daytime conditions and headlight exposure under nighttime conditions degrades driver visibility and remains a persistent challenge for safe roadway operation, yet the mechanisms governing outcome differentiation under glare conditions across complex roadway, vehicle, and driver interactions remain insufficiently understood.

  Predicting Glare-Related Traffic Outcomes with Transformer-Based Explainable Tabular Deep Learning · 2026 · DOI

• This is due to the difficulty in generalizing from sparse data to dense data. On the other hand, models trained without pretraining, including the randomly initialized GPT-2, exhibit substantially higher errors, indicating that having a similar architecture alone is insufficient without pretrained knowledge.

  DG-LLM: Decomposition-based dynamic graph adaptation of large language models for spatiotemporal traffic forecasting · 2026 · DOI

• Condition (A5) relies on universal approximation theorems (Hornik 1991 for bounded σ, Kidger-Lyons 2020 for ReLU), but the paper notes the ReLU result applies to L²(Rd; Rd) rather than the required L²(μ1) space; rigorous verification that this approximation property holds for all measure distributions μ1 in the problem is missing.

  A deep learning approach to multi-marginal optimal transport via Hilbert space embeddings of probability measures · 2026 · DOI

• The theoretical conditions (A1)-(A5) require absolute continuity with respect to Lebesgue measure and specific kernel properties (Gaussian, Matérn), but experimental validation is limited to synthetic data; performance on discrete, singular, or empirical measures from real applications needs investigation.

  A deep learning approach to multi-marginal optimal transport via Hilbert space embeddings of probability measures · 2026 · DOI

• Memory consumption scales severely with dimensionality in the experiments (Tables 10, 12: 32GB+ for d=10 and N=10 marginals), but the paper provides no analysis of computational complexity bounds or memory requirements as functions of dimension d and number of marginals N for the deep learning optimal transport algorithm.

  A deep learning approach to multi-marginal optimal transport via Hilbert space embeddings of probability measures · 2026 · DOI

• The convergence proof assumes that the number of hidden layer dimensions D can be arbitrarily chosen during theoretical analysis, but the practical algorithm in Section 3 fixes D values; the relationship between theoretical requirements on D and empirical performance with fixed architecture dimensions needs explicit investigation.

  A deep learning approach to multi-marginal optimal transport via Hilbert space embeddings of probability measures · 2026 · DOI

• Theorem 1 guarantees asymptotic satisfaction of marginal constraints via MMD penalty and weak convergence of transport maps, but the convergence of the transport cost value itself is not addressed and remains an open direction for further analysis in the multi-marginal optimal transport framework.

  A deep learning approach to multi-marginal optimal transport via Hilbert space embeddings of probability measures · 2026 · DOI

• The acoustic module relies on Mel-Filterbank Energy features extracted by ESP32-S3 hardware, but there is no evaluation of how siren acoustic signatures vary across different emergency vehicle types (ambulances, fire trucks, police vehicles) or how geographic regional siren standards affect detection performance in international deployments.

  False Positive Reduction in Emergency Vehicle Detection Using a Multimodal Edge-Based System · 2026 · DOI

• The paper does not compare the proposed asymmetric multimodal architecture against continuous or parallel multimodal fusion approaches in terms of precision, recall, and computational trade-offs under equivalent high-noise urban conditions. A systematic comparison framework evaluating different fusion topologies for emergency vehicle detection on edge devices is absent.

  False Positive Reduction in Emergency Vehicle Detection Using a Multimodal Edge-Based System · 2026 · DOI

• The system's asymmetric AND-gate fusion architecture was validated over only an 11-day field study with 1,380 detection events in a single deployment location. Scalability and generalization of the multimodal edge-based system across diverse geographic regions, traffic patterns, and varying siren acoustic characteristics in different countries remain unvalidated.

  False Positive Reduction in Emergency Vehicle Detection Using a Multimodal Edge-Based System · 2026 · DOI

• Six of the nine missed detections resulted from visual occlusion where emergency vehicles were partially or fully blocked from the camera field of view. The current single-camera YOLOv5-based vision module needs enhancement through multi-camera coverage strategies or trajectory prediction algorithms to handle occluded vehicle scenarios in real-world traffic.

  False Positive Reduction in Emergency Vehicle Detection Using a Multimodal Edge-Based System · 2026 · DOI

• The multimodal emergency vehicle detection system failed to detect nine events due to acoustic interference in high-noise urban environments where siren frequencies were masked by overlapping sound sources. Future work should develop noise-robust audio preprocessing techniques specifically for siren detection in traffic-congested settings with competing acoustic signals.

  False Positive Reduction in Emergency Vehicle Detection Using a Multimodal Edge-Based System · 2026 · DOI

• The paper does not evaluate how model performance degrades under specific real-world traffic conditions (e.g., rush hours vs. off-peak periods, adverse weather events, accidents, special events) or how the trained models transfer across different expressways with different traffic characteristics.

  Multimodal traffic flow analysis and congestion prediction on expressways: evaluating the importance of machine learning models and features for improving prediction accuracy in urban traffic management · 2026 · DOI

• The logistic regression model achieves perfect or near-perfect metrics (accuracy 0.9993, AUC 1.0000) which suggests potential data leakage, overfitting on the training set, or unrealistic dataset characteristics; the paper does not investigate these anomalies or validate LR performance on independent, temporally distinct traffic expressway datasets.

  Multimodal traffic flow analysis and congestion prediction on expressways: evaluating the importance of machine learning models and features for improving prediction accuracy in urban traffic management · 2026 · DOI

• While RF achieved optimal accuracy-efficiency balance for real-time implementation in urban traffic management, the paper does not evaluate RF's performance under actual real-time deployment conditions with streaming traffic flow data or test its scalability across multiple expressway corridors simultaneously.

  Multimodal traffic flow analysis and congestion prediction on expressways: evaluating the importance of machine learning models and features for improving prediction accuracy in urban traffic management · 2026 · DOI

• Future research will explore adaptive DP mechanisms that dynamically adjust noise levels based on data sensitivity, model convergence, or specific client contributions, aiming for greater efficiency and tighter privacy guarantees with minimal accuracy sacrifice. Future work could focus on developing XAI techniques specific to federated MoE models to understand how different experts contribute to predictions for specific cities or scenarios, and how DP impacts interpretability. • Exploring Advanced Federated Personalization Strategies: While MoE provides adaptation, more explicit personalized federated learning (pFL) techniques, such as FedMeta, FedAvgM, or client clustering, could be investigated.

  Deep heterogeneity learning for cross-city transit forecasting: a differentially private federated framework with mixture-of-experts and seasonal decomposition · 2026 · DOI

• However, despite advancements in deep RL algorithms like DDPG, PPO, SAC, and advanced CV techniques such as semantic segmentation and multi-camera tracking, the direct integration of these technologies remains underexplored in many domains.

  A Review of Deep Multi-Objective Reinforcement Learning and Vision-Based Systems for Smart Cities · 2025 · DOI

• The results suggest comparable or slightly better short-term forecasting performance under the evaluated dataset and protocol, while the conclusions remain dataset-specific and should be further examined on additional traffic networks.

  U-GRU for Short-Term Urban Traffic Speed Forecasting with Ordered-Node Feature Transformation and Channel–Temporal Recalibration · 2026 · DOI

• Cheng, “Forecasting urban traffic states with sparse data using hankel temporal matrix factorization,” INFORMS Journal on Computing, 2024.

  Calibrating adaptive smoothing methods for freeway traffic reconstruction · 2026 · DOI

• An underexplored challenge is inductive forecasting: predicting O-D flows for new or unobserved locations, which is essential for evaluating network expansions and new facility placements.

  Inductive Transportation Origin-Destination Demand Forecasting via Bayesian Destination-Choice Modeling · 2026 · DOI

• At present, people's livelihood demand data collection is often affected by factors such as limited geographical coverage, a high proportion of sudden demands, and inconsistent data recording periods.

  People’s livelihood demand prediction based on improved generative adversarial networks and reinforcement learning · 2026 · DOI

• • A model-agnostic meta-learning framework that leverages data from multiple cities to estimate Macroscopic Fundamental Diagrams (MFDs) in networks with scarce data from loop detectors.

  Learning to learn the macroscopic fundamental diagram using physics-informed and model agnostic machine learning · 2026 · DOI

• Thus, a solid understanding of the difficulties in predicting congestion is required because the transportation system varies widely between non-congested and congested states.

  Optimizing Traffic Flow in Smart Cities: Soft GRU-Based Recurrent Neural Networks for Enhanced Congestion Prediction Using Deep Learning · 2023 · DOI

Most-cited papers in Traffic Prediction and Management Techniques

• Sensing Data Supported Traffic Flow Prediction via Denoising Schemes and ANN: A Comparison · IEEE Sensors Journal · 2020 · 269 citations
• A deep learning based traffic crash severity prediction framework · Accident Analysis & Prevention · 2021 · 204 citations
• Predicting real-time traffic conflicts using deep learning · Accident Analysis & Prevention · 2020 · 197 citations
• Comprehensive systematic review of information fusion methods in smart cities and urban environments · Information Fusion · 2024 · 193 citations
• An analytic framework using deep learning for prediction of traffic accident injury severity based on contributing factors · Accident Analysis & Prevention · 2021 · 160 citations
• Traffic management approaches using machine learning and deep learning techniques: A survey · Engineering Applications of Artificial Intelligence · 2024 · 139 citations
• TrafficGPT: Viewing, processing and interacting with traffic foundation models · Transport Policy · 2024 · 138 citations
• A Resource-Aware Multi-Graph Neural Network for Urban Traffic Flow Prediction in Multi-Access Edge Computing Systems · IEEE Transactions on Consumer Electronics · 2024 · 130 citations
• UrbanGPT: Spatio-Temporal Large Language Models · 2024 · 129 citations
• A deep learning method based on an attention mechanism for wireless network traffic prediction · Ad Hoc Networks · 2020 · 121 citations

Most recent work

• Deep heterogeneity learning for cross-city transit forecasting: a differentially private federated framework with mixture-of-experts and seasonal decomposition · Frontiers in Future Transportation · 2026
• Traffic flow prediction based on spatiotemporal feature fusion and Graph Convolutional Network · Engineering Computations · 2026
• Multimodal traffic flow analysis and congestion prediction on expressways: evaluating the importance of machine learning models and features for improving prediction accuracy in urban traffic management · Journal of Engineering and Applied Science · 2026
• Public Transport Crowd Estimation via Bus Stop Images Using AI · INTERNATIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT · 2026
• False Positive Reduction in Emergency Vehicle Detection Using a Multimodal Edge-Based System · Engineering Technology & Applied Science Research · 2026
• A deep learning approach to multi-marginal optimal transport via Hilbert space embeddings of probability measures · Statistics and Computing · 2026
• Road Traffic Accident Severity Prediction Using Machine Learning · INTERNATIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT · 2026
• Intelligent traffic routing in urban cities using a data-driven machine learning framework for real-time route optimization in Kampala, Uganda · Discover Applied Sciences · 2026
• Cross-State Traffic Accident Risk Prediction Using Self-Supervised Representation Learning on the US-Accidents Dataset · The Saudi Journal of Applied Sciences and Technology · 2026
• Sensing the Road Ahead: Advanced Randomized Algorithm for Situational Awareness in Smart Transportation · International Journal of Computational Intelligence and Applications · 2026

Related topics in Engineering