Project summary, final thoughts, and next steps
This project successfully built a data-driven regression model to predict the resale price of used Ford vehicles, achieving an R² of 0.840 and RMSE of £1,900.
Using 17,966 real market listings from the United Kingdom, we demonstrated that Linear Regression with One-Hot Encoding provides an excellent balance of interpretability, speed, and accuracy for vehicle price prediction.
Achieved R² of 0.840, surpassing the 0.80 goal with a model that explains 84% of price variance.
Created 22 professional visualizations covering distributions, correlations, outliers, and category-level analysis.
5-Fold cross-validation with std < 0.01 confirms the model generalises well and doesn't overfit.
Identified that year and mileage are the top price drivers, with clear implications for dealers and sellers.
Complete reproducible pipeline with < 100ms training and < 1ms inference times.
Built this interactive portfolio website with automatic notebook extraction and glassmorphism design.
This project demonstrates the power of data-driven decision making in the automotive industry. By leveraging machine learning, we can provide consistent, objective price estimates that remove human bias and guesswork.
While Linear Regression may seem simple compared to modern deep learning approaches, it proved to be the right tool for this problem — offering interpretability, speed, and excellent performance. Sometimes the simplest solution is the best one.
The project also highlights the importance of feature engineering — our choice of One-Hot vs Label Encoding resulted in an 11 percentage point difference in R². Small decisions can have big impacts.
Ford Used Car Dataset
Kaggle
17,966 used Ford vehicle listings from the United Kingdom
Thank you to the Kaggle community for providing the dataset, the open-source community for the excellent Python libraries, and to you for exploring this project.
Let's connect and discuss machine learning, data science, or potential opportunities
If you found this project helpful or interesting, please consider giving it a star on GitHub. Your support is greatly appreciated!