Abstract:
"In the development of mobility-on-demand systems and traveler information systems,
the ability to predict travel times is critical. Riders and drivers who use such systems
benefit from accurate travel time estimation as it aids in the decision-making
processes.
In this research paper, a meta learning model is introduced to predict the static travel
time of trip trajectories using Catboost and XGBoost as base models and Linear
Regression as the meta learner to efficiently learn from the predictions of the base
models and provide an accurate travel time prediction. The Meta-predictor model is
evaluated and compared with many regression models where the model performs
impressively. The prediction values of the model have reduced the Mean Absolute
Error (MAE) and Root Mean Squared Error (RMSE) while improving the correlation
when compared to actual travel time of trips.
The author of the research demonstrates that the accuracy of static travel time
prediction could be improved by combining better performing regression models with
a Meta learner which helps to minimize the errors of prediction values compared to
the actual trip travel time when applied to large scale data"
