Developed a predictive model using machine learning algorithms to forecast Airbnb prices based on various features such as location, amenities, and property type.
- Python
- scikit-learn
- pandas
- osmnx
- geopandas
- Handling missing data
- Feature engineering
- Model evaluation
- Random Forest Regressor
- Linear Regression
- Geospatial Models (SLX & SAR)
- Key factors influencing property prices include location, property size, and proximity to attractions.
- Spatial influences also play a role in price predictions.
- The dots represent Airbnb listings in Prague. Brighter colors indicate higher prices.
- As expected, the highest prices are concentrated in the Old Town of Prague.
| Model | Property Features | POI Features | Spatial Lag | Spatial Cross Correlation | Relative Improvement RMSE |
|---|---|---|---|---|---|
| Ordinary Least Squares Regression | β | β | β | β | 0% |
| Geospatial Regression | β | β | β | β | 8% |
| Random Forest | β | β | β | β | -2% |
| Random Forest (with POIs) | β | β | β | β | 14% |
- Points of Interest (POIs) are locations that may attract people, such as restaurants, bars, and public transportation.
- Spatial Lag: Measures the influence of neighboring properties' prices on the price of a property. It captures spatial autocorrelation of property prices.
- Spatial Cross Correlation: Measures the relationship between the spatial distribution of different variables, capturing spatial dependence between features.
- Effect of Space: Considering spatial influences improves price predictions.
- Best Models: The best models are either linear models with complex spatial features (spatial regression) or non-linear models (random forest) with simpler geospatial features (POIs).