Enhancing Expressiveness of Models for Static Route-Free Estimation of Time of Arrival in Urban Environments

Schleibaum, Sören; Müller, Jörg P.; Sester, Monika

doi:10.1016/j.trpro.2022.02.054

Cited by 4 publications

(22 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As regards SC3, we observe that the reported feature importances for the temperature are low or close to zero. While this could indicate that the ETA models have not learned the underlying pattern, similarly to Schleibaum, Müller, and Sester [SMS22], we argue that the overall feature contribution of the weather or temperature is low.…”

Section: First-level Model Explanationmentioning

confidence: 47%

“…We additionally include the Washington DC dataset to increase the generalizability of our experiment as regards the usage of ensembles to increase the prediction precision. For both datasets, we rely on the feature engineering described by Schleibaum, Müller, and Sester [SMS22], which makes use of or enhances the dataset by the following features: the location-based ones with (1) the pickup/dropoff as degree-based coordinates and (2) the indices of a 50 meter square grid as an alternative representation. To represent the start time of a trip, (3) the month, (4) week, (5) weekday, and (6) the indices of a 5-minute time-bin, which represents the hour and minute, are used.…”

Section: Datasetmentioning

confidence: 99%

“…Outlier Removal. For removing outliers, we also use the criteria from Schleibaum, Müller, and Sester [SMS22] and the description of the following method partly reproduces their wording. Overall, around three percent of the trips from the New York City dataset and around 19 percent from the Washington DC dataset are filtered out.…”

Section: Datasetmentioning

confidence: 99%

“…We take the three ETA models proposed by Schleibaum, Müller, and Sester [SMS22] and their hyperparameters as our first-level models; we choose these because, with a bagging-, a boosting-and a neural network-based model, they provide a good basis for a heterogeneous ensemble. As alternatives for the second-level model, we consider the same machine learning methods and add a relatively simple Multiple Linear Regression (MLR).…”

Section: Estimated Time Of Arrival Modelsmentioning

confidence: 99%

“…We take the three ETA models proposed by Schleibaum, Müller, and Sester [SMS22] as our first-level models as well as their hyperparameters, which have been chosen via Bayesian optimization. The first model is an RF-based one (L1-RF) with 300 trees and a maximum tree depth of 89; the number of maximal features per node is chosen automatically and the minimum number of samples per leaf and split are set to four.…”

Section: Models For Estimating the Time Of Arrivalmentioning

confidence: 99%

See 4 more Smart Citations

An Explainable Stacked Ensemble Model for Static Route-Free Estimation of Time of Arrival

Schleibaum¹,

Müller²,

Sester³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

To compare alternative taxi schedules and to compute them, as well as to provide insights into an upcoming taxi trip to drivers and passengers, the duration of a trip or its Estimated Time of Arrival (ETA) is predicted. To reach a high prediction precision, machine learning models for ETA are state of the art. One yet unexploited option to further increase prediction precision is to combine multiple ETA models into an ensemble. While an increase of prediction precision is likely, the main drawback is that the predictions made by such an ensemble become less transparent due to the sophisticated ensemble architecture. One option to remedy this drawback is to apply eXplainable Artificial Intelligence (XAI). The contribution of this paper is three-fold. First, we combine multiple machine learning models from our previous work for ETA into a two-level ensemble model-a stacked ensemble model-which on its own is novel; therefore, we can outperform previous state-ofthe-art static route-free ETA approaches. Second, we apply existing XAI methods to explain the firstand second-level models of the ensemble. Third, we propose three joining methods for combining the first-level explanations with the second-level ones. Those joining methods enable us to explain stacked ensembles for regression tasks. An experimental evaluation shows that the ETA models correctly learned the importance of those input features driving the prediction.

show abstract

Section: First-level Model Explanationmentioning

confidence: 47%

Section: Datasetmentioning

confidence: 99%

Section: Datasetmentioning

confidence: 99%

Section: Estimated Time Of Arrival Modelsmentioning

confidence: 99%

Section: Models For Estimating the Time Of Arrivalmentioning

confidence: 99%

See 3 more Smart Citations