03 / MACHINE LEARNING · DATA
Used Car Price Prediction
From messy data to meaningful estimates.
RANDOM FOREST REGRESSIONMODEL / 150 TREES
85.4%TEST R²
01RAW DATA
02CLEANING
03FEATURES
04RANDOM FOREST
05PREDICTION
4,300+ records5-fold cross-validation
01 / OVERVIEW
An end-to-end regression pipeline that transforms used-car records into contextual price predictions.
THE PROBLEM
Used-car pricing depends on interacting features, and raw records contain skewed distributions and outliers that can distort a model.
THE SOLUTION
A structured preprocessing pipeline applies log transformations, feature engineering and percentile-based outlier removal before fitting a Random Forest regressor. An interactive interface makes the resulting predictions accessible.
4,300+RECORDS PROCESSED
85.4%TEST R²
150RANDOM FOREST TREES
5-foldCROSS-VALIDATION
02 / ARCHITECTURE
How the pieces connect.
- 014,300+ raw records↓
- 02Cleaning + feature engineering↓
- 03150-tree Random Forest↓
- 04Evaluation + prediction UI↓
03 / CAPABILITIES
Built into the product.
- Preprocessing and feature engineering↗
- Log transformations and outlier removal↗
- 150-tree Random Forest regression↗
- 5-fold cross-validation↗
- Interactive prediction interface↗
- Input validation and contextual price visualisation↗
ENGINEERING CHALLENGES
Preparing skewed data for regression, evaluating model performance beyond the training data and validating inputs to the prediction interface.
WHAT I LEARNED
The project highlights how preprocessing and feature engineering shape model quality. Five-fold cross-validation complements the held-out test result.