Details
| Original language | English |
|---|---|
| Pages (from-to) | 432-441 |
| Number of pages | 10 |
| Journal | Transportation Research Procedia |
| Volume | 62 |
| Early online date | 11 Mar 2022 |
| Publication status | Published - 2022 |
| Event | 24th Euro Working Group on Transportation Meeting, EWGT 2021 - Aveiro, Portugal Duration: 8 Sept 2021 → 10 Sept 2021 |
Abstract
Scheduling of taxis can reduce cost and potentially decreases CO2emissions. However, with a rising number of taxis or travel requests, the time for computing schedules increases. A promising alternative is to estimate trip durations based on historical trip data without calculating routes. Based on an analysis of the state of the art, in this paper we identify and investigate two limitations of route-free Estimated Time of Arrival (ETA) models: First, the overall set of features considered by state of-the-art models is limited. For instance, some potential relevant features (such as weather-related ones) are not considered at all. Also, different models use different sets of features, such as the linear distance between pickup and dropoff location, in diverse and partly inconsistent ways. For those features generally considered, we find different representations, e.g., for trip start time. Second, while discretization of degree-based coordinates for pickup/dropoff locations via spatial binning is very common in state-of-the-art ETA models, the chosen grid cell sizes vary widely and apparently arbitrarily. The contribution of this paper is threefold: First, we propose to enhance route-free ETA models by additional features and investigate the influence of the feature representation on the prediction precision based on a benchmark dataset. Second, we compare different grid cell topologies and sizes as regards their effect on the prediction precision of ETA. Third, we construct and evaluate three types of Machine Learning (ML) models. Our findings indicate that the results outperform state-of-the-art static route-free ETA estimation models.
Keywords
- Estimated time of arrival, machine learning, taxi fleet management, travel time estimation
ASJC Scopus subject areas
- Social Sciences(all)
- Transportation
Cite this
- Standard
- Harvard
- Apa
- Vancouver
- BibTeX
- RIS
In: Transportation Research Procedia, Vol. 62, 2022, p. 432-441.
Research output: Contribution to journal › Conference article › Research › peer review
}
TY - JOUR
T1 - Enhancing Expressiveness of Models for Static Route-Free Estimation of Time of Arrival in Urban Environments
AU - Schleibaum, Sören
AU - Müller, Jörg P.
AU - Sester, Monika
PY - 2022
Y1 - 2022
N2 - Scheduling of taxis can reduce cost and potentially decreases CO2emissions. However, with a rising number of taxis or travel requests, the time for computing schedules increases. A promising alternative is to estimate trip durations based on historical trip data without calculating routes. Based on an analysis of the state of the art, in this paper we identify and investigate two limitations of route-free Estimated Time of Arrival (ETA) models: First, the overall set of features considered by state of-the-art models is limited. For instance, some potential relevant features (such as weather-related ones) are not considered at all. Also, different models use different sets of features, such as the linear distance between pickup and dropoff location, in diverse and partly inconsistent ways. For those features generally considered, we find different representations, e.g., for trip start time. Second, while discretization of degree-based coordinates for pickup/dropoff locations via spatial binning is very common in state-of-the-art ETA models, the chosen grid cell sizes vary widely and apparently arbitrarily. The contribution of this paper is threefold: First, we propose to enhance route-free ETA models by additional features and investigate the influence of the feature representation on the prediction precision based on a benchmark dataset. Second, we compare different grid cell topologies and sizes as regards their effect on the prediction precision of ETA. Third, we construct and evaluate three types of Machine Learning (ML) models. Our findings indicate that the results outperform state-of-the-art static route-free ETA estimation models.
AB - Scheduling of taxis can reduce cost and potentially decreases CO2emissions. However, with a rising number of taxis or travel requests, the time for computing schedules increases. A promising alternative is to estimate trip durations based on historical trip data without calculating routes. Based on an analysis of the state of the art, in this paper we identify and investigate two limitations of route-free Estimated Time of Arrival (ETA) models: First, the overall set of features considered by state of-the-art models is limited. For instance, some potential relevant features (such as weather-related ones) are not considered at all. Also, different models use different sets of features, such as the linear distance between pickup and dropoff location, in diverse and partly inconsistent ways. For those features generally considered, we find different representations, e.g., for trip start time. Second, while discretization of degree-based coordinates for pickup/dropoff locations via spatial binning is very common in state-of-the-art ETA models, the chosen grid cell sizes vary widely and apparently arbitrarily. The contribution of this paper is threefold: First, we propose to enhance route-free ETA models by additional features and investigate the influence of the feature representation on the prediction precision based on a benchmark dataset. Second, we compare different grid cell topologies and sizes as regards their effect on the prediction precision of ETA. Third, we construct and evaluate three types of Machine Learning (ML) models. Our findings indicate that the results outperform state-of-the-art static route-free ETA estimation models.
KW - Estimated time of arrival
KW - machine learning
KW - taxi fleet management
KW - travel time estimation
UR - http://www.scopus.com/inward/record.url?scp=85127058188&partnerID=8YFLogxK
U2 - 10.1016/j.trpro.2022.02.054
DO - 10.1016/j.trpro.2022.02.054
M3 - Conference article
AN - SCOPUS:85127058188
VL - 62
SP - 432
EP - 441
JO - Transportation Research Procedia
JF - Transportation Research Procedia
SN - 2352-1457
T2 - 24th Euro Working Group on Transportation Meeting, EWGT 2021
Y2 - 8 September 2021 through 10 September 2021
ER -