AutoGauge for Used Vehicle Price Estimation and Visual Damage Assessment

Used vehicle valuation requires both a price estimate from vehicle attributes and an interpretable account of visible condition. This paper presents AutoGauge, a web application integrating XGBoost regression, YOLOv8n damage detection, and SHapley Additive exPlanations. A dataset audit identifies 7,396 car records after excluding 7,313 two-wheelers from the supplied raw file. In the recorded tabular benchmark, XGBoost achieves a mean absolute error of 43,754.48 price units, a root mean squared error of 83,611.20, and an R² of 0.9189 on 1,480 test records. Its mean absolute error is 7.35% lower than Random Forest on this split and 12.31% lower in subsequent five-fold training-partition cross-validation. A separate damage evaluation on 374 images reports mAP50 of 0.7269 and mAP50–95 of 0.5662. The application translates detections into a heuristic condition score and uses a price model trained with simulated condition adjustments. Because price records and damage images are unpaired, these component results do not demonstrate that image fusion improves observed-price prediction. Preprocessing leakage, historical data coverage, and uncalibrated condition penalties are discussed as limits of the present evidence.