Predicting Trust In Autonomous Vehicles: Modeling Young Adult Psychosocial Traits, Risk-Benefit Attitudes, And Driving Factors With Machine Learning

Research Background and Issues

• Identified Problems or Challenges:
  The authors identified a lack of trust in autonomous vehicles (AVs) among young people as a major barrier to the widespread adoption of autonomous driving technology. Despite the potential benefits of AVs, such as reducing traffic accidents, improving efficiency, and lowering environmental impact, public distrust of the technology has hindered its adoption. This issue of trust is particularly pronounced in high-risk scenarios involving personal safety, such as driving.

• Significance:
  Trust is a prerequisite for the active acceptance and use of technology. Even if AVs demonstrate superior performance and safety, the absence of user trust could lead to adoption failure. Therefore, understanding the formation of trust and its influencing mechanisms is critical for the successful deployment of this technology.

• Research Motivation and Related Work:
  While existing studies have explored trust in AVs through factors such as personality traits, attitudes, and experiences, most research has focused on a limited set of variables. This narrow approach has restricted the understanding of interactions between these variables. Furthermore, most studies target general populations rather than specific groups, making it difficult to address the needs of particular user demographics. To address these gaps, the authors aim to conduct a comprehensive survey of variables and apply machine learning methods, with a specific focus on young adults, who are expected to be key users of future transportation technologies.

Proposed Solution

• Proposed Methods or Solutions:
  This study collected data from young adults (sample size = 1,457) through a questionnaire covering a wide range of variables, including psychological factors, driving behaviors, technological attitudes, and perceptions of AV risks and benefits. Machine learning techniques (e.g., random forests, support vector machines) were then used to predict user trust, with SHAP (Shapley Additive exPlanations) employed to interpret the key influencing factors.

• Innovations:

  1. The study investigates a broader range of factors than any previous research, encompassing diverse personal characteristics and evaluations of AV-related risks and benefits.
  2. Machine learning methods are utilized to capture complex nonlinear relationships between features, which traditional statistical methods might overlook.
  3. Explainable AI techniques (SHAP) are applied to reveal the relative importance of different variables and their contributions to trust prediction.
  4. By focusing on young adults, the study provides in-depth insights into this specific demographic group.

• Implementation Steps:

  1. Data Collection: Comprehensive personal characteristics and attitudes were gathered through a questionnaire, including driving behaviors, psychological factors, and perceptions of AV risks and benefits.
  2. Model Development: Machine learning algorithms were used to predict whether users belong to "high-trust" or "low-trust" groups.
  3. Feature Importance Analysis: SHAP was used to interpret the model results, ranking the importance of variables and their impact on trust prediction.
  4. Isolated Variable Analysis: Feature ablation experiments (removing certain variables) were conducted to further explore the influence of specific variables on the model.

Research Findings

• Specific Outcomes Achieved:

  1. The machine learning model achieved an accuracy of 85.8%, effectively predicting young adults' trust in AVs.
  2. Analysis of factor importance revealed that perceptions of risks and benefits were critical for trust prediction, with "overall risk and benefit assessment" being the most significant contributor.
  3. Psychological, driving, and technology-related characteristics were found to be less important than expected, challenging previous assumptions about trust mechanisms.
  4. "Humanized" AV decision-making models were shown to be associated with higher levels of trust.

• Advantages:

  1. The approach demonstrated greater predictive power than traditional methods like linear regression or variance analysis, capturing nonlinear and interaction effects.
  2. The findings provide actionable design guidelines for improving communication and design of AV technologies targeted at young adults.

• Experimental or Evaluation Results:

  • SHAP analysis revealed that "overall risk and benefit assessment," "accident reduction," and "ease of use" were the primary factors influencing trust ratings. In contrast, complex psychological or driving personality traits were less significant for trust prediction.
  • Feature ablation experiments confirmed the importance of risk and benefit perceptions. A model using only risk and benefit features achieved nearly the same predictive accuracy as the full model (84.6%).

• Limitations and Future Directions:

  1. Limitations: The study relied on self-reported data, which may introduce subjective bias. Additionally, the sample primarily consisted of young university students in the United States, limiting the generalizability of the findings.
  2. Future Directions:
     • Replicate the study across different age groups and cultural contexts to validate the generalizability of the results.
     • Use behavioral trust data instead of self-reported data to improve the precision of trust prediction.
     • Test the practical application of the study's recommendations in AV design to verify their effectiveness in enhancing trust.

Conclusion

This study demonstrates that young adults' trust in AVs can be effectively predicted based on personal characteristics and reveals that perceptions of risks and benefits are the most critical factors for trust prediction. The findings provide important insights for the design and policymaking of future AV technologies, including strategies for creating transparent and user-friendly systems to enhance trust. Additionally, the study serves as a model for personalized design and research for other demographic groups.