01
The Problem.
Real estate transaction data is publicly available but difficult to exploit — large volumes, missing values, inconsistent formats. 13.3M raw transactions reduced to 3.7M exploitable rows after filtering. Making accurate decisions about housing prices and market trends requires significant data cleaning and enrichment before it becomes actionable.
02
The Solution.
A full data pipeline that collects, cleans, and enriches 3.7M public real estate transactions to identify market trends and estimate property values. A Random Forest model predicts housing prices based on features such as surface area, number of rooms, location, and property type — reaching 87% precision.
89.9%
Precision
Random Forest score
8
Specialized Models
In production
3.7M
Transactions
DVF 2022–2025
The Stack.
01
Data & ML
Python · pandas · NumPy · scikit-learn · joblib · API BAN · INSEE Filosofi · DVF
02
Backend
Flask · Flask-CORS · Lazy model loading
03
Frontend
HTML5 · CSS3 · JavaScript ES6 · Three.js · Fetch API
04
Outils
Jupyter Notebook · VS Code · Git · GitHub