🇪🇸 Versión en Español

BAIRES Project 📈

Analysis and prediction of the real-estate market in Buenos Aires

Click here to go to the notebook

ℹ️ Description

The project can be considered at three stages.

1. First, preprocessing. The dataset is cleaned and prepared (Droping columns and rows with missing or null values, calulating and imputing data wisely to not generate a bias in the model then).
2. Then, price and location of properties in Ciudad Autonoma de Buenos Aires (CABA) is analysed. Cheapest and most expensive neighborhood are identified. Also, it is shown that real locations and published locations mismatch a lot of times, and most of times published location correspond with more expensive areas.
3. At stage 3 remaining preprocessing is done and binaries are created. Then two models are trained: DecisionTreeRegressor and KNeighbours. A test set was curated for final evaluation, while train set was partitioned in several manners for evaluation: with simple train_test_split, and with Kfold implementation. GridSearch was used for optimization. And root mean square error (RMSE) was adopted as error measure.

Finally, the models can predict price of properties with an estimated error of +/-$ 40000 (depending de model)

Skills: Python, Scikit-learn, Matplotlib, Seaborn, Crossvalidation, DecisionTreeRegressor, KNeighbours, Pandas, Numpy

Datos provided by Properati

🔭 Dataset attributes

• id: property id
• created_on: when property was uploaded to the website
• operation: rent or sell)
• property_type: house, department, store, PH
• place_with_parent_names: Where is located the property, according to the publisher
• lat-lon: coordinates
• lat: latitude
• lon: longitude
• price_aprox_usd: aproximated price in usd
• surface_total_in_m2: Total surface (in square meters)
• surface_covered_in_m2: Covered surface (in square meters)
• price_usd_per_m2: price per square meter in usd (price / surface)
• floor: floor number (if appropriate)
• rooms: number of rooms
• expenses: exenses (if appropriate)
• barrio: neighborhood in agrreement with official charts.
• properati_url: url of publication in Properati

➡️ This project was developed as an activity of the ACAMICA DATA SCIENCE course.