Modeling and prediction of driving performance measures based on multi-output convolutional Gaussian process

Area of Science:

• Automotive Engineering
• Machine Learning
• Data Science

Background:

• Driving performance measures (DPMs) are crucial for vehicle safety and assessing driver behavior.
• Acquiring DPM data is costly and limited, necessitating accurate prediction models for unobserved conditions.
• Existing DPM prediction models often fail to account for the complex interactions among different DPMs.

Purpose of the Study:

• To propose a novel method for modeling and predicting DPMs that explicitly incorporates inter-DPM interactions.
• To develop a flexible and interpretable modeling framework for DPM prediction.
• To demonstrate the superiority of the proposed method over existing benchmark approaches.

Main Methods:

• A multi-output convolutional Gaussian process (MCGP) model was developed.
• The MCGP model was designed to capture the interactions between different DPMs.
• The proposed method was evaluated against three benchmark methods using a DPM dataset.

Main Results:

• The MCGP method demonstrated superior performance in DPM modeling and prediction compared to benchmark methods.
• The results confirmed the effectiveness of incorporating inter-DPM interactions for improved prediction accuracy.
• The proposed method offers both modeling flexibility and an interpretable structure for DPM interactions.

Conclusions:

• The novel MCGP method provides a significant advancement in DPM prediction by effectively modeling inter-DPM dependencies.
• This approach enhances the reliability of DPM predictions, supporting the development of advanced driver assistance systems.
• The findings highlight the importance of considering DPM interactions for robust driving behavior analysis and safety applications.