Automation of Leasing Vehicle Return Assessment Using Deep Learning Models

The vehicle damage assessment includes classifying damage and estimating its repair cost and is an essential process in vehicle leasing and insurance industries. It contributes heavily to the actual cost the customer has to pay. The standard practices follow manual identification of damages and cost estimation of repairs, resulting in noisy images of the damaged parts, inconsistent categorization of damage types, and high variance in repair costs estimation between two appraisers. We employ explainable machine learning to highlight how the standard ML models and their training protocols fail when dealing with a dataset acquired without a standard procedure. In this paper, we present a multi-task image regression model for the leasing vehicle return assessment that leverages the car configuration to reduce the cost of repair assessment. Our solution achieves a 50% error reduction in the repair cost estimates. Furthermore, we present remedies base on hierarchical taxonomy and cost-sensitive loss to improve the damage classification accuracy.