Construction and Validation of a Risk Annotation Model for Logistics Claim Settlement Based on Stratified Quantile Regression

Abstract

Against the backdrop of flourishing digital economies and e-commerce, the logistics sector's pursuit of high-quality development demands increasingly sophisticated and intelligent claims handling services. To accurately identify claims risk in logistics waybills, optimise post-sales costs, and enhance customer satisfaction, this paper constructs a stratified quantile regression risk annotation model based on historical waybill data from a logistics enterprise. The model centres on two core metrics: actual payout amount and claim discrepancy. First, data cleansing and preprocessing are conducted, with claim differences calculated and visually analysed. Subsequently, waybills were grouped into 11 descending tiers based on actual payout amounts, ensuring sufficient sample size within each tier. Within each tier, linear interpolation determined 85th and 97th percentile thresholds, categorising waybills into three risk levels: "reasonable claims", "excessive claims", and "severely excessive claims". Validated through ANOVA testing and coefficient of variation analysis, this model demonstrates significant stratification efficacy. It effectively captures claim risk disparities across varying payout scales, providing scientific rationale and practical solutions for logistics enterprises' claims risk management.