Project: Predicting Airbnb Prices with Machine Learning

📝 Description

Developed a predictive model using machine learning algorithms to forecast Airbnb prices based on various features such as location, amenities, and property type.

🛠️ Technologies Used

• Python
• scikit-learn
• pandas
• osmnx
• geopandas

🚀 Challenges

• Handling missing data
• Feature engineering
• Model evaluation

🤖 Models Used

• Random Forest Regressor
• Linear Regression
• Geospatial Models (SLX & SAR)

📊 Findings

• Key factors influencing property prices include location, property size, and proximity to attractions.
• Spatial influences also play a role in price predictions.

🌍 Map of Airbnb Prices in Prague

• 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.

📈 Results and Model Comparison

Model	Property Features	POI Features	Spatial Lag	Spatial Cross Correlation	Relative Improvement RMSE
Ordinary Least Squares Regression	✅	❌	❌	❌	0%
Geospatial Regression	✅	✅	✅	✅	8%
Random Forest	✅	❌	❌	❌	-2%
Random Forest (with POIs)	✅	✅	❌	❌	14%

📍 What are Points of Interest (POIs)?

• Points of Interest (POIs) are locations that may attract people, such as restaurants, bars, and public transportation.

🗺️ What are Spatial Lags and Spatial Cross Correlation?

• 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.

🔑 Key Findings

• 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).