Travel Time Prediction and Explanation with Spatio-Temporal Features: A Comparative Study

Abstract

Travel time information is used as input or auxiliary data for tasks such as dynamic navigation, infrastructure planning, congestion control, and accident detection. Various data-driven Travel Time Prediction (TTP) methods have been proposed in recent years. One of the most challenging tasks in TTP is developing and selecting the most appropriate prediction algorithm. The existing studies that empirically compare different TTP models only use a few models with specific features. Moreover, there is a lack of research on explaining TTPs made by black-box models. Such explanations can help to tune and apply TTP methods successfully. To fill these gaps in the current TTP literature, using three data sets, we compare three types of TTP methods (ensemble tree-based learning, deep neural networks, and hybrid models) and ten different prediction algorithms overall. Furthermore, we apply XAI (Explainable Artificial Intelligence) methods (SHAP and LIME) to understand and interpret models’ predictions. The prediction accuracy and reliability for all models are evaluated and compared. We observed that the ensemble learning methods, i.e., XGBoost and LightGBM, are the best performing models over the three data sets, and XAI methods can adequately explain how various spatial and temporal features influence travel time.

On the Soundness of XAI in Prognostics and Health Management (PHM)

“…Several studies have applied XAI methods to regression time series. For instance, Ahmed et al [13] used SHAP and LIME to explain the predictions of a model trained to forecast travel time. In another study, Vijayan [14] employed a deep learning multi-output regression model to predict the relationship between optical design parameters of an asymmetric Twin Elliptical Core Photonic Crystal Fiber (TEC-PCF) and its sensing performances.…”

Developing Prediction Model of Travel Times of the Logistics Fleets of Large Convenience Store Chains Using Machine Learning

“…Concerning studies related to travel time prediction, Wang et al [23] proposed a wide deep recurrent learning model, which comprised train-wide models, deep neural networks, and recurrent neural networks, to estimate the travel time for a fixed route and departure time. Ahmed et al [24] proposed an explainable artificial intelligence method to evaluate and compare different prediction models and to explain the effects of temporal and spatial characteristics on travel time. Servos et al [25] used extremely randomized trees, adaptive boosting, and support vector regression (SVR) to solve problems with a small quantity of data in a short time.…”

Geospatial XAI: A Review

“…The next two major categories are research in traffic and transport and research in human-related use cases. For the use cases in traffic and transport, the goals were to predict car accidents [11], predict travel times [28] or demand for ride-sourcing [29], or tune and test machine learning models with travel data [30,31]. The category of human-related use cases covers a wide range of research, but it directly affects people's lives.…”

“…Extreme GBM [11,20,24,28,29,31,33,34,[36][37][38], Light GBM [20,25,38], Gradient BM [11,29], Explainable BM [13], Natural GB [20], Adaptive Boosting [20] Neural Networks 17 Artificial ANN/Multilayer perceptron [12,18,22,26,27,29,35,38,40], Deep NN [14,28,37], Convolutional NN [21,26,39], Recurrent NN [32], Superposable NN [23] Tree-based Models 15 Random Forest [12,13,15,16,[18][19][20]27,29,37,38,40] Decision Tree [11,29,…”

“…Deep NN [14,28,37], Convolutional NN [21,26,39], Recurrent NN [32], Superposable NN [23] Tree The most commonly used machine learning models were boosting approaches, neural networks, and tree-based models. Although boosting approaches are also based on treebased models, they are placed in a separate category due to the difference in the construction of these trees compared to single decision trees or random forests.…”