XGBoost-based risk prediction model for massive vehicle recalls using consumer complaints

Area of Science:

• Automotive Engineering
• Risk Management
• Data Science

Background:

• Vehicle recalls pose significant safety risks and economic burdens.
• Predicting recall risk is challenging due to complex data and evolving factors.
• Existing methods often overlook structured complaint data and distinct recall stages.

Purpose of the Study:

• To develop high-precision predictive models for vehicle recall risk.
• To identify key risk factors influencing vehicle recalls using consumer complaints.
• To enhance proactive risk management strategies for automakers and regulatory bodies.

Main Methods:

• Employed the XGBoost machine learning model for in-depth analysis.
• Utilized comprehensive data from the National Highway Traffic Safety Administration (NHTSA).
• Integrated structured consumer complaint data and official recall records.

Main Results:

• Achieved exceptional model performance across various time windows.
• Demonstrated high predictive accuracy and stability with area under the curve values up to 18 months.
• Distinguished between indicators for initial and subsequent recalls, addressing different vehicle lifecycle stages.

Conclusions:

• The XGBoost model provides valuable support for proactive vehicle recall risk management.
• Systematic integration of structured complaint data enhances recall prediction accuracy.
• The study bridges a critical gap in predicting recall risk at different stages of a vehicle's lifecycle.