House Price Prediction

House Price Prediction is a four-person team project that estimates property prices from both what a listing says and what it looks like. The team scraped and processed a multimodal dataset (tabular features + images) from arenda.az, separating apartments and houses.

The tabular baseline uses an ExtraTreesRegressor, reaching R² = 0.82 on apartments and 0.75 on houses. Listing photos are then encoded with CLIP and fused with the tabular features through an MLP, producing a multimodal pipeline that scores R² = 0.73 on houses and 0.72 on apartments.

The trained model is deployed behind a FastAPI service for real-time predictions.