Model Automatization — Car Sales
Overview
Developed as a team machine learning project to accurately estimate the market valuation of used SUV vehicles across 22,377 listings. We leveraged Microsoft Azure Machine Learning to automate data ingestion, experiment tracking, and distributed hyperparameter tuning on dedicated cloud compute clusters.
Key Features
- Cloud Pipeline on Azure ML: Configured Azure ML Datasets, Environments, and Compute Clusters for distributed model training and reproducibility.
- Model Exploration & Selection: Benchmarked four distinct regression algorithms: K-Nearest Neighbors (KNN), Linear Regression, Extra Trees Regressor, and Random Forest.
- Hyperparameter Optimization: Utilized GridSearchCV across multi-node clusters to tune tree depth, min samples split, and feature subsets.
- Price Driver Feature Importance: Extracted feature importance metrics identifying vehicle mileage, model year, and engine specifications as primary drivers of resale value.
Architecture & Implementation
Raw vehicle listings were preprocessed using automated scikit-learn pipelines with outlier removal, missing value imputation, and categorical one-hot encoding. Azure ML Jobs executed parallel hyperparameter sweeps across cluster nodes. Evaluation metrics (RMSE, MAE, R²) were logged and visualized in Azure ML Studio.
Results & Performance
Random Forest achieved the best performance on unseen test data with an RMSE of 7656.09 and MAE of 4784.67, providing an automated valuation engine that estimates used SUV prices with an average error of ~4,785 monetary units.