AutoWorth-Ai Car price prediction
تفاصيل العمل
Buying a used car usually comes with one core question: Is this actually a fair price, am I overpaying or what is the actual market price of the car?
I built AutoWorth AI to solve this problem—predicting the fair market value of a vehicle and advising buyers on whether the seller’s asking price represents a good deal.
Working with the 100,000 UK Used Car Dataset, I built the entire pipeline from scratch:
Data Acquisition & Combination ➔ Data Cleaning ➔ EDA ➔ Feature Engineering ➔ Preprocessing ➔ Model Training ➔ Evaluation ➔ Hyperparameter Tuning ➔ Final Model Selection
I trained and evaluated several regression models (Linear Regression, Decision Tree, Random Forest, Gradient Boosting and XGBoost) using R², MAE, and RMSE then choosing the best model while also checking for overfitting and analyzing feature importance.
Smart Deal Advisor:The project goes beyond price prediction by including an Advisor that compares the predicted market price with the seller’s price and classifies it from Great Deal 🟢 to Overpriced 🔴.
Interactive Web App (Streamlit): Deployed the pipeline into an intuitive Streamlit interface, where users can enter the car details, get the predicted market price, and receive the deal recommendation directly through the interface.
This project was a great opportunity to understand that Machine Learning isn't simply fit and predict the target, The real learning was in understanding the data, deciding how to preprocess it, choosing the right approach, evaluating the results, investigating overfitting, and understanding why the model behaves the way it does.
Special appreciation to our instructor, Eng. Fayrouz Ahmed, for making the training practical, challenging, and focused on understanding what happens behind the code. Her guidance encouraged us to experiment, question our results, and become more confident in building ML solutions independently.
My appreciation also goes to Telecom Egypt and Alexandria Training Center -WE for the opportunity and organization, CyberGuardX Training Academy for their support throughout the program.
Explore the full code and repository on GitHub: