A Review of Traffic State Prediction (TSP) Methods in Intelligent Transportation Systems (ITS)

Ahanin, Fatemeh; Mustapha, Norwati; Zolkepli, Maslina; Husin, Nor Azura · 2023 · Crossref

DOI: 10.6007/ijarbss/v13-i3/16683

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This review paper addresses the critical challenge of traffic congestion in metropolitan areas, which negatively impacts mobility, travel time, and quality of life. The authors focus on Traffic State Prediction (TSP) as a pivotal component of Intelligent Transportation Systems (ITS) for mitigating these issues. The motivation stems from the need for accurate, real-time traffic data—such as flow, velocity, and density—to support effective traffic management. The paper reviews existing methodologies for TSP, categorizing them into three main Artificial Intelligence approaches: Probabilistic Reasoning, Shallow Machine Learning (SML), and Deep Learning (DL). It also examines data collection methods, contrasting traditional static sensors (e.g., inductive loops) with Floating Car Data (FCD) derived from cellular networks, noting that while FCD offers cost-effective coverage, it suffers from GPS errors and data sparsity. The review analyzes specific algorithms within each category. Under Probabilistic Reasoning, it covers Fuzzy Logic, Hidden Markov Models (HMM), Gaussian Processes (GPs), and Bayesian Networks (BN). For instance, HMMs are noted for modeling time-series data with MAE accuracies around 6–10%, while GPs are useful for reducing feature numbers and estimating location errors. Shallow Machine Learning methods include Artificial Neural Networks (ANNs), Regression Models, Decision Trees, and Support Vector Machines (SVMs). The authors highlight that ANNs, particularly BackPropagation Neural Networks, offer efficient forecasting with low processing times, whereas Linear Regression often yields lower accuracy compared to ensemble methods like Random Forest. Decision Trees are criticized for producing binary results unsuitable for nuanced congestion prediction. Deep Learning is presented as the dominant method for TSP due to its ability to automatically extract spatiotemporal features. The paper details Convolutional Neural Networks (CNNs), which treat traffic data as time-space matrices to capture temporal and spatial dependencies, and Recurrent Neural Networks (RNNs), specifically Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) variants. LSTM models are frequently cited for achieving high performance in speed prediction, with some studies reporting MAE values as low as 1.45. Hybrid models, such as those combining Graph Neural Networks with RNNs, are also discussed for modeling complex road network relationships. The significance of this review lies in its comprehensive synthesis of AI applications in ITS, providing a roadmap for researchers. The authors conclude that while current methods are effective, future research should focus on integrating multi-source data, such as social media updates and weather conditions, to enhance prediction accuracy. They emphasize that precise short-term speed prediction is essential for providing drivers with real-time guidance, thereby improving safety and reducing congestion. The paper serves as a foundational resource for understanding the evolution and current state of machine learning techniques in traffic management.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• traffic density