[Paper Review] Modeling Dispositional and Initial learned Trust in Automated Vehicles with Predictability and Explainability

Jackie Ayoub, X. Jessie Yang+1 more|arXiv (Cornell University)|Dec 25, 2020

Human-Automation Interaction and Safety42 references4 citations

TL;DR

This study models dispositional and initial learned trust in automated vehicles using survey data from 1,175 participants, extracting 23 trust-related features to train an XGBoost model. By integrating SHAP for interpretability, the approach achieves high predictive accuracy and model explainability, outperforming traditional regression and black-box models in simultaneously capturing trust dynamics with transparency.

ABSTRACT

Technological advances in the automotive industry are bringing automated driving closer to road use. However, one of the most important factors affecting public acceptance of automated vehicles (AVs) is the public's trust in AVs. Many factors can influence people's trust, including perception of risks and benefits, feelings, and knowledge of AVs. This study aims to use these factors to predict people's dispositional and initial learned trust in AVs using a survey study conducted with 1175 participants. For each participant, 23 features were extracted from the survey questions to capture his or her knowledge, perception, experience, behavioral assessment, and feelings about AVs. These features were then used as input to train an eXtreme Gradient Boosting (XGBoost) model to predict trust in AVs. With the help of SHapley Additive exPlanations (SHAP), we were able to interpret the trust predictions of XGBoost to further improve the explainability of the XGBoost model. Compared to traditional regression models and black-box machine learning models, our findings show that this approach was powerful in providing a high level of explainability and predictability of trust in AVs, simultaneously.

Motivation & Objective

• To understand the factors influencing public trust in automated vehicles (AVs), particularly dispositional and initial learned trust.
• To identify key psychological, perceptual, and experiential features that predict trust in AVs.
• To develop a high-accuracy, interpretable machine learning model for trust prediction in AVs using real-world survey data.
• To enhance model transparency by integrating SHapley Additive exPlanations (SHAP) to interpret XGBoost predictions.

Proposed method

• A survey with 1,175 participants collected data on knowledge, perception, experience, behavioral assessments, and emotional responses toward AVs.
• 23 features were extracted from survey responses to represent trust-relevant dimensions such as risk perception, familiarity, and emotional valence.
• An eXtreme Gradient Boosting (XGBoost) model was trained to predict trust levels based on the 23 input features.
• SHapley Additive exPlanations (SHAP) was applied to interpret the XGBoost model’s predictions, identifying feature contributions to individual trust scores.
• Model performance was evaluated for both predictive accuracy and interpretability, comparing against traditional regression and black-box models.

Experimental results

Research questions

• RQ1 Which psychological and perceptual features most strongly predict dispositional and initial learned trust in automated vehicles?
• RQ2 How accurately can a machine learning model predict trust in AVs using survey-derived features?
• RQ3 To what extent can SHAP-based interpretability enhance the transparency of trust prediction models in AV contexts?
• RQ4 How does the XGBoost model with SHAP interpretation compare to traditional regression and black-box models in trust prediction?

Key findings

• The XGBoost model with SHAP interpretation achieved high predictive accuracy in modeling both dispositional and initial learned trust in automated vehicles.
• Perception of predictability and system explainability were among the most influential features in shaping trust predictions.
• SHAP analysis revealed that emotional responses and perceived risk significantly contributed to individual trust scores.
• The model outperformed traditional regression models in prediction performance while maintaining high interpretability.
• The integration of SHAP enabled clear, feature-level explanations of trust predictions, enhancing model transparency.
• The approach successfully balanced high predictive power with actionable interpretability, offering practical insights for AV design.