Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Efficient and Interpretable Traffic Destination Prediction using Explainable Boosting Machines

Yasin Yousif, Jörg Müller

TL;DR

The paper tackles traffic destination prediction for autonomous driving by adopting Explainable Boosting Machines (EBMs), a glass-box variant of Generalized Additive Models, and evaluating them on SDD, InD, and Argoverse. It introduces a multi-modal prediction framework in which multiple mode-specific EBMs are trained, and mode probabilities are computed by combining mode-specific and main unimodal log-likelihoods, enabling a flexible selection of the top predictions. The approach yields interpretable results through feature importance and partial dependence graphs while achieving competitive performance on pedestrian-dominated datasets and offering insight into interactions between features. However, performance on road-context-rich Argoverse is weaker, highlighting the need for road-network processing and per-type modeling; the work also emphasizes reproducibility by providing training code and supplementary material.

Abstract

Developing accurate models for traffic trajectory predictions is crucial for achieving fully autonomous driving. Various deep neural network models have been employed to address this challenge, but their black-box nature hinders transparency and debugging capabilities in a deployed system. Glass-box models offer a solution by providing full interpretability through methods like \ac{GAM}. In this study, we evaluate an efficient additive model called \ac{EBM} for traffic prediction on three popular mixed traffic datasets: \ac{SDD}, \ac{InD}, and Argoverse. Our results show that the \ac{EBM} models perform competitively in predicting pedestrian destinations within \ac{SDD} and \ac{InD} while providing modest predictions for vehicle-dominant Argoverse dataset. Additionally, our transparent trained models allow us to analyse feature importance and interactions, as well as provide qualitative examples of predictions explanation. The full training code will be made public upon publication.

Efficient and Interpretable Traffic Destination Prediction using Explainable Boosting Machines

TL;DR

The paper tackles traffic destination prediction for autonomous driving by adopting Explainable Boosting Machines (EBMs), a glass-box variant of Generalized Additive Models, and evaluating them on SDD, InD, and Argoverse. It introduces a multi-modal prediction framework in which multiple mode-specific EBMs are trained, and mode probabilities are computed by combining mode-specific and main unimodal log-likelihoods, enabling a flexible selection of the top predictions. The approach yields interpretable results through feature importance and partial dependence graphs while achieving competitive performance on pedestrian-dominated datasets and offering insight into interactions between features. However, performance on road-context-rich Argoverse is weaker, highlighting the need for road-network processing and per-type modeling; the work also emphasizes reproducibility by providing training code and supplementary material.

Abstract

Developing accurate models for traffic trajectory predictions is crucial for achieving fully autonomous driving. Various deep neural network models have been employed to address this challenge, but their black-box nature hinders transparency and debugging capabilities in a deployed system. Glass-box models offer a solution by providing full interpretability through methods like \ac{GAM}. In this study, we evaluate an efficient additive model called \ac{EBM} for traffic prediction on three popular mixed traffic datasets: \ac{SDD}, \ac{InD}, and Argoverse. Our results show that the \ac{EBM} models perform competitively in predicting pedestrian destinations within \ac{SDD} and \ac{InD} while providing modest predictions for vehicle-dominant Argoverse dataset. Additionally, our transparent trained models allow us to analyse feature importance and interactions, as well as provide qualitative examples of predictions explanation. The full training code will be made public upon publication.
Paper Structure (16 sections, 3 equations, 16 figures, 2 tables)

This paper contains 16 sections, 3 equations, 16 figures, 2 tables.

Table of Contents

  1. Introduction
  2. GAM Formula
  3. Methodology
  4. Multiple Modes Output
  5. Outputs Probabilities
  6. SDD
  7. InD
  8. Argoverse
  9. Preprocessing Steps
  10. Road Network Data
  11. Defining Modes
  12. Experimental Evaluation
  13. Conclusion
  14. Representing Nearby Traffic
  15. EBM Hyperparameters
  16. ...and 1 more sections

Figures (16)

  • Figure 1: Left: full training set of target variable. Right: target variable set is split into multiple clusters.
  • Figure 2: How the road network is represented (Left). Each mode rectangle (Right) return the geometric center of the drivable area underneath it. Red and green colors represent importance for x and y respectively
  • Figure 3: SDD: Global Feature Average Importance for X and Y
  • Figure 4: SDD: Partial Dependence Graph on X and Y axis for the best 6 features
  • Figure 5: InD: Global Feature Average Importance for X and Y
  • ...and 11 more figures