How It Works

Learn how our predictive AI analyzes vehicle data to estimate the best market price.

Prediction Pipeline

📝

1. Input Data

You provide details like brand, model, year, and transmission.

🔍

2. Condition Analysis

The system calculates deductions based on 10 physical factors.

⚙️

3. AI Valuation

Our predictive AI matches your inputs with real-world market patterns.

💵

4. Price Estimation

The app generates a price estimate and a detailed deviation analysis.

AI Prediction System

PriceMyCar uses Predictive AI technology powered by the HistGradientBoostingRegressor (HGBR) algorithm. This model was trained on thousands of real-world used car sales transactions to capture non-linear depreciation trends—for example, how the value of a luxury vehicle drops much faster than an economy family car as mileage increases.

Why HGBR was selected:

HGBR groups continuous numerical variables (like mileage or age) into 256 integer bins (histogram-based binning). This significantly reduces computation time, improves noise resistance, and acts as a strong regularizer. We applied L2 regularization (2.0) and restricted tree complexity (max leaf nodes = 25) to prevent overfitting.

Key Parameters Evaluated by AI:

  • Brand & Model: Popularity and demand representation via Frequency Encoding.
  • Car Age: Derived feature (2025 - year) capturing the primary driver of value loss.
  • Mileage (Odometer): Cumulative physical wear and tear.
  • Fuel & Transmission: Encoded preferences (manual vs automatic and fuel types).
  • Ownership History: Encoded ordinal tiers (First Owner to Test Drive Car).
  • Interaction Terms: Engineered feature (age × km) capturing cumulative depreciation.

AI Valuation Metrics

AI Model Accuracy (R²)
The proportion of used car price variance successfully explained by our machine learning features on the held-out test split.
80.6%
Average Price Deviation (MAE)
The average absolute nominal error of predictions relative to actual market transactions (equivalent to 107,252 INR).
Rp 20.1M
Generalization Gap
The difference between Train R² and Test R² (86.8% vs 80.6%). A low gap indicates the model is highly stable and does not overfit.
6.2%