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

Abstract

This study employed the XGBoost model to conduct an in-depth analysis of consumer complaints and identified the key risk factors predicting vehicle recalls in the US market, providing valuable proactive risk management support for automakers and regulatory agencies. We leveraged the extensive data resources from National Highway Traffic Safety Administration to construct high-precision recall risk prediction models to predict the risk of recall. The models exhibited exceptional performance across different time windows, particularly maintaining a high level of area under the curve values over a prediction timespan of up to 18 months, demonstrating their predictive accuracy and stability. Our study contributes to risk management theory by addressing the challenges of integrating consumer complaints into predictive models for vehicle recall risk. While prior research has primarily focused on text mining of complaint content, our work systematically incorporates structured complaint data and recall records to enhance predictive accuracy. Also, our research distinguishes the indicators for the initial recall after launch to the market and the indicators for subsequent recalls, and bridges a critical gap in recall risk prediction at different stages of a vehicle's life cycle.