Answers are grounded in this paper and cite their sources.
Passenger car unit (PCU) values are needed to express mixed traffic in a common unit for capacity and level-of-service analysis. This study summarises grade-specific PCU values for five vehicle classes on three multilane highways in Egypt and examines how well machine-learning models reproduce them. PCU values were computed with the speed–area (Chandra) method from the mean speeds of each class on grades between −3% and +3%, giving 135 observations (3 roads × 9 grades × 5 classes). An artificial neural network (ANN) was trained on six inputs (vehicle class, grade, annual average daily traffic, class share, class mean speed and projected area). Because the training target is the Chandra PCU, the ANN is treated explicitly as a surrogate of that method. Its accuracy was compared with eight alternatives, including a vehicle-class mean baseline, linear and log-linear regression, support vector regression, random forest, gradient boosting, k-nearest neighbours and a smaller ANN, under random 10-fold, leave-one-grade-out and leave-one-road-out cross-validation. Mean PCU increased from the steepest downgrade to the steepest upgrade by 6–9% for buses (3.30 to 3.59), light commercial vehicles (2.43 to 2.59) and heavy vehicles (5.07 to 5.39), while two-wheelers remained at about 0.13. Within the studied roads, the ANN reproduced the Chandra values closely (root mean square error, RMSE, 0.033 PCU in random cross-validation and 0.039 when whole grade levels were withheld). When an entire road was withheld, the ANN error increased about eight times (RMSE 0.251), and no model was clearly better than the vehicle-class mean (RMSE 0.185). Vehicle class and speed accounted for most of the predictive information. The results support the use of such surrogates for interpolation within observed conditions, but not for transfer to other roads without data from additional sites.
No takes yet. Share an insight, caveat, or question.
Ramadan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: