Application of artificial neural networks for operating speed prediction at horizontal curves: a case study in Egypt

Semeida, Ahmed Mohamed · 2014 · OpenAlex-citations

DOI: 10.1007/s40534-014-0033-3

archive: archived pipeline: cataloged verified

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

Summary

This study addresses the lack of operating speed prediction models for multi-lane highways, particularly in Egypt, where previous research has predominantly focused on rural two-lane roads. Horizontal curves significantly influence vehicle speeds, yet fewer studies have analyzed the relationship between geometric features and operating speeds on multi-lane highways. The research aims to develop and compare prediction models for the 85th-percentile operating speed (V85) of cars and trucks using both conventional regression analysis and artificial neural networks (ANNs). The methodology involved collecting field data from 78 horizontal curve sections across four multi-lane highways in Egypt, categorized into agricultural and desert highways. Data collection included geometric variables such as lane width, shoulder width, median width, curve radius, deflection angle, curve length, and superelevation. Spot speed data for passenger cars and trucks were gathered under free-flow conditions using radar guns, with approximately 18,000 total speed observations. The study employed two analytical approaches: multiple linear regression with stepwise selection to identify significant variables, and multi-layer perceptron (MLP) neural networks trained using backpropagation. The dataset was split into training (70–90%) and testing (10–30%) subsets to evaluate model performance and prevent overfitting. The results indicate that ANN models provided superior prediction accuracy compared to regression models. For cars, the curve radius was identified as the most influential variable, while for trucks, the median width was the most significant factor. The best-performing ANN model for cars achieved an overall R² of 0.932 and a root mean square error (RMSE) of 5.77 km/h. The ANN model for trucks demonstrated even higher accuracy, with an overall R² of 0.95 and an RMSE of 4.33 km/h. In contrast, the best regression models yielded lower R² values (0.892 for cars and 0.88 for trucks) and higher RMSE values. The study also validated that the derived models were statistically significant and conceptually reasonable, with variable signs aligning with engineering judgment. The significance of this research lies in providing reliable, high-accuracy tools for predicting operating speeds on multi-lane highways, which are crucial for road safety and geometric design. By demonstrating the superiority of ANNs over traditional regression methods, the study suggests that non-linear modeling techniques should be considered for complex traffic engineering problems. The findings offer specific insights into how different geometric features affect cars and trucks differently, aiding planners and designers in optimizing highway geometry for improved safety and efficiency.

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.

• road geometry