# Dynamic Toll Prediction Using Historical Data on Toll Roads: Case Study of the I-66 Inner Beltway

arXiv: https://arxiv.org/abs/2104.10684 · HTML: https://mlsanity.com/paper/2104.10684/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 6 of its 51 references that are in the map.
- First posted on arXiv 20 Apr 2021.

## Builds on (most cited first, 6 of 6)

- [Adam: A Method for Stochastic Optimization](https://mlsanity.com/paper/1412.6980/) (2014)
- [Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift](https://mlsanity.com/paper/1502.03167/) (2015)
- [Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting](https://mlsanity.com/paper/1506.04214/) (2015)
- [LSTM: A Search Space Odyssey](https://mlsanity.com/paper/1503.04069/) (2015)
- [Estimating Historical Hourly Traffic Volumes via Machine Learning and Vehicle Probe Data: A Maryland Case Study](https://mlsanity.com/paper/1711.00721/) (2017)
- [A Matching Mechanism with Anticipatory Tolls for Congestion Pricing](https://mlsanity.com/paper/2002.05269/) (2020)

## Abstract (arXiv)

Providing the users of a dynamic tolling system with predictions of tolling prices and the travel time difference between the toll road and the alternative routes enables them to make their travel decisions before starting their trip. This study aims to provide accurate predictions of tolling price through training and testing random forest, multilayer perceptron, and long short-term memory models and compare them with the current situation that the best prediction is extending the current toll to the next timesteps. The prediction time horizon includes five 6-minute time intervals ahead of the present time. The prediction performance of models over the testing set reveals that while all the models were significantly better than the base model, the random forest outperforms all models. For instance, while in the trained models, the mean absolute error range is from 1.5 to 2.5 for the next six minutes to the next 30 minutes, respectively, the same measure in the base model is in the range of 2.5 to 6. The prediction of travel time difference along the toll road and its alternative route with the shortest travel time revealed that the multilayer perceptron performs marginally better than the base model. However, due to a relatively stable travel time difference, the current travel time difference is an acceptable prediction for the next 30 minutes prediction horizon.