CarPricePred

CarPricePred is a predictive analytics project developed to solve a common problem in the used car market: valuation uncertainty. Developed in cooperation with a colleague, this tool uses a neural network to estimate car values based on key parameters.

The Origin

The project stemmed from a real need to understand fair pricing in the used car market. Instead of manually checking listings, we aimed to build software that could autonomously calculate the correct price based on Brand, Model, and Year of Production.

Data & Scraping

To build a robust dataset, we engineered an ethical web scraper targeting AutoScout24. After processing approximately 300,000 listings, we cleaned and consolidated a database of 160,000 valid records using Pandas and NumPy. This phase involved significant data cleaning, outlier removal, and feature engineering.

The Neural Network

The core is a TensorFlow neural network. We employed techniques like Robust Scaler for mileage and year, and One-Hot Encoding for categorical variables. The model was trained to recognize patterns across ~2,000 car models, achieving an accuracy close to 95% in final testing phases.