This paper examines the transferability of the Safety Performance Function (SPF) of the Highway Safety Manual (HSM) and other 10 international SPFs for total crashes on rural multi-lane divided roads in Egypt. Four segmentation approaches are assessed in the transferability of the international SPFs, namely: (1) one-kilometer segments (S1); (2) homogenous sections (S2); (3) variable segments with respect to the presence of curvatures (S3); and (4) variable segments with respect to the presence of both curvatures and U-turns (S4). The Mean Absolute Deviation (MAD), Mean Prediction Bias (MPB), Mean Absolute Percentage Error (MAPE), Pearson χ² statistic, and Z-score parameters are used to evaluate the performance of the transferred models. The overdispersion parameter (k) for each transferred model and each segmentation approach is recalibrated using the local data by the maximum likelihood method. Before estimating the transferability calibration factor (Cr), three methods were used to adjust the local crash prediction of the transferred models, namely: (1) the HSM default crash modification factors (CMFs); (2) local CMFs; and (3) recalibrating the constant term of the transferred model. The latter method is found to outperform the first two methods. Besides, the results show that the segmentation method would affect the performance of the transferability process. Moreover, the Italian SPFs based on the S1 segmentation method outperforms the HSM and all of the investigated international SPFs for transferring their models to the Egyptian rural roads.

This study re-evaluates Egypt’s Differential Speed Limit (DSL) policy by proposing Lane-Based Speed Limits (LBSL), an innovative strategy where speed limits vary across highway lanes. To address a critical research gap, we developed a microsimulation framework using Simulation of Urban Mobility (SUMO) with parameters specifically justified for Egyptian driver behavior. While constrained by data availability, our model moves beyond generic defaults by integrating a key stand-still gap parameter (CC0 = 1.5 m) calibrated from previous research on Egyptian conditions, ensuring a more accurate representation of local driving patterns. We simulated three-lane highway scenarios under varying traffic volumes (500–1750 veh/h/lane), heavy vehicle percentages (2.5–20%), and speed limits (100, 110, 120 km/h). This compared LBSL configurations (speed reductions of 10, 15, and 20 km/h between lanes) against Uniform Speed Limit (USL) and current DSL strategies. Performance was assessed through average delay, travel speed, lane-changing frequency, and conflict ratios using Time-to-Collision thresholds (≤ 2.50, ≤ 1.50, and ≤ 0.50 s). The results indicate that Egypt’s existing DSL strategy offers no statistically significant safety advantages over USL, questioning its effectiveness. Conversely, LBSL significantly improved safety, reducing conflicts by up to 60%, though it introduced a trade-off with operational efficiency by increasing delays by 71–870%. The primary contribution is a contextually grounded analysis demonstrating LBSL as a safer strategic alternative to Egypt’s incumbent DSL policy. This provides Egyptian policymakers with evidence-based insights for potential traffic safety reforms and offers a methodological framework for adapting microsimulation to other regions with similar data constraints.

This paper explores the attitudes and travel mode preferences of Mansoura University students towards eco-friendly transportation options, specifically shared electric scooters (SESs) and shared bikes (SBs). An online stated preference survey was administered to 790 students via Google Forms, yielding approximately 9,590 choice scenarios for analysis. The dataset was split into training (70%) and testing (30%) sets, with 6,717 cases in the training set and 2,880 cases in the testing set. A multinomial logit model and ten machine learning classifiers were used to predict mode choice behavior. The variables considered in the models include total travel time, travel cost, gender, health issues, bike ownership, ability to ride a bike and scooter, residency, prior knowledge of shared modes, smartphone ownership, frequency of public transport use, trip cost, trip time, and monthly personal income. Model performance was evaluated at two levels: individual and aggregate. At the individual level, accuracy reflects the overall percentage of correct predictions. At the aggregate level, accuracy is assessed based on the predicted market shares for each mode. The Extreme Gradient Boosting (XGB) model was the most accurate, achieving an accuracy slightly above 60%, followed by the Gradient Boosting (GB) and Random Forest (RF) models, both with accuracies slightly above 50%. Ensemble models like RF, GB, and XGB provided relatively accurate aggregate predictions with minimal deviations from actual mode shares. These models slightly overestimated the bus mode share but offered accurate predictions for other modes, making them robust choices for this classification task.

Traffic simulation models require calibration to capture local driving behavior, particularly in developing countries where conditions differ from Western defaults. This study calibrates three car-following models—Krauss, Wiedemann 99 (W99), and Intelligent Driver Model (IDM)—with lane-change parameters for heterogeneous traffic on six two-lane two-way roads in Egypt’s Nile Delta. A sensitivity analysis reduced eight lane-change parameters to three critical ones: lcOpposite, lcSpeedGain, and lcStrategic. A Genetic Algorithm then optimized car-following and lane-change parameters jointly across multiple seeds and generations. The best configurations achieved mean absolute percentage errors of 8.79% (IDM), 8.94% (W99), and 10.56% (Krauss), indicating that IDM and W99 performed similarly under the original four-metric validation, while Krauss was less accurate overall and exceeded the overtaking-rate threshold. suggesting calibration methodology matters more than model choice. Minimum standstill gap (minGap) treatment varied by model: freeing it improved IDM and Krauss performance, while fixing it at 1.50 m benefited W99. Constraining minGap in IDM increased error by 2.06 percentage points and altered parameter sensitivity, with lcStrategic becoming dominant and tau showing reduced sensitivity. Notably, SUMO’s uncalibrated Krauss defaults achieved 10.11% error under understaturated conditions, though this should not extend to congested scenarios. Simulated standard deviations underestimated observed variability, primarily due to aggregation mismatches rather than model inadequacy.