Vehicle price estimation using Machine Learning on web-scraped second-hand market data

Intending to sell my uncommonly-trimmed and low-mileage MY2010 FIAT Panda 4x4 MultiJet Rossignol Edition on the Swiss second-hand car market, I “scratched my own itch” and wrote software to help me price my vehicle.

To this end, I first developed a set of Python scripts for web-scraping structured data on second-hand vehicles from Swiss site autoscout24.ch. Then, I utilized various Machine Learning algorithms implemented in Weka for clustering and regression on the scraped data, generated various submodels, and generated ensemble estimates for pricing.

At the request of a friend who also wanted to sell his car and buy a different one through AutoScout24, I performed the same analysis for different brands and models, and together we developed possible business models for data-driven SaaS offerings based on periodically scraping sub-datasets to evaluate the “aging” of online car offers and possibly determine opportunities for car-flipping.

No public-facing service was ever launched, as the web-scraped data was not my own IP; however, the system provided insights that were used for the sale of at least one vehicle :)