Pricing recommendation: MAPE 25% — bad for sellers

Ситуация: Real estate platform recommends asking price. MAPE = 25% (predicted vs actual sold). Sellers complain. Improve.

Pricing model: comparable sales, square meters, location, condition. 10 input features.

Доступные данные

• listings: id, asking_price, sold_price (если sold), features
• comparable_sales: source_listing_id, comp_id, distance, time_diff
• model_predictions: listing_id, predicted_price, model_version

Задачи

• MAPE decomposition: by region, price band, time-on-market.
• Feature analysis: что drives error.
• Comparison set quality.
• Improvement plan.

Все кейсы для подготовки → ← Все кейсы