How TruPrice Calculates Used Car Fair Value

Updated July 2026

A transparent look at the XGBoost model behind TruPrice — what data it uses, how it handles condition and mileage, and why it produces a different number than listings-based averages.

Stage 1: XGBoost Asking Price Prediction

The core model is an XGBoost gradient boosting regressor trained on real Pakistan used car market transaction data. XGBoost is a tree-based ensemble method that handles the non-linear relationships in car pricing — for example, a 2020 Corolla Altis depreciates differently than a 2020 Cultus for the same calendar year.

Stage 2: Negotiation Buffer

The model outputs an estimated asking price — the price a seller would list the car for. A category- and age-specific negotiation buffer (2D lookup table: category × age bracket) converts this to a base transaction price. This represents what informed buyers actually pay.

Stage 3: Condition-Based Inspection Discounts

When inspection scores are provided, each of the 8 sections is compared to a par score for the car's category and age bracket. The deviation from par determines a condition band (Excellent / Good / Fair / Bad). Each band maps to a discount rate applied to the base price. The sum of all section discounts produces the final fair price.

Stage 4: Mileage Verdict

A second XGBoost model predicts expected mileage for the car's make, model, age bracket, and category. Actual mileage entered is compared to this prediction. The resulting deviation produces a verdict: Low Mileage, Normal Mileage, or High Mileage — shown alongside the fair price for context.

Coverage

Frequently Asked Questions