Airbnb, Inc. is an American San Francisco-based company operating an online marketplace for short-term homestays and experiences. The company acts as a broker and charges a commission from each booking. The company not only have revolutionized the tourism industry but has also facilitated an unaffordable increase in home rents and a lack of regulation (https://en.wikipedia.org/wiki/Airbnb).
With this in mind, the objectives of this project were:
- Visualise the properties offered in the 20 Arrondissements of Paris during Quarters Q2 - Q4 of 2022
- Find features that impact the price of a listing
- Identify potential inactive listing with outlier prices - noise
- Identify the most expensive and cheapest neighbourhoods
- Identyfy the type of accommodations and properties more often offered by Airbnb
- Identify listings that offered for long-term stays
- Processing the data to predict the listing price per night with Machine Learning algorithms
- Optimize the best model hyperparameters
- Best model explainability - XAI Shapley number computation
Airbnb datasets Q2-Q4 2022 can be sourced from http://insideairbnb.com/get-the-data/
- pandas
- numpy
- matplotlib
- sklearn
- xgboost
- lightgbm
- skorch
- missigno
- joblib
- shap
GIS libraries
- geopandas
- contextily
- folium
Frameworks
- PyTorch
-
Interactive map available for
airbnb_paris.ipynbhttps://ace-aitech.github.io
- 75 % of the listing has minimum_nights up to 4.
- There are 7232 listings in airbnb Paris that required a minimun_nights stay of 30 nights covering the 8.74% unique listings from Q2-Q4 2022.
- The most common property_type is rental unit
- Most common type of room Entire home/apt
Note The listings were recategorised in short and long-term if the minimum_nights was >=30
- The most expensive room type is Hotel room
- The cheapest accommodation is Shared room
- Most expensive properties are floor and Villa
- Cheapest neighbourhoods Ménilmontant $101.787, Buttes-Chaumont $116.75
- The most expensive neighbourhoods are Élysée $260.112, Louvre $257.11
There are 7232 listings in airbnb Paris that required a minimun_nights stay of 30 nights covering the 8.74% unique listings from Q2-Q4 2022.
| amenity | frequency |
|---|---|
| wifi | 88783 |
| kitchen | 85683 |
| essentials | 83260 |
| heating | 83255 |
| long term stays allowed | 77452 |
| smoke alarm | 74465 |
| hot water | 73666 |
| hair dryer | 73414 |
| dishes and silverware | 71057 |
| washer | 70354 |
The dataset after wrangling, cleaning and encoding contained 215 features. SelectPercentile with mutual_info_regression was used to select only the 50 percentile. The table below only shows the top 10 features.
| Feature | score |
|---|---|
| latitude | 0.432559 |
| longitude | 0.432547 |
| accommodates | 0.218494 |
| reviews_per_month | 0.158682 |
| private_bathroom | 0.157154 |
| bedrooms | 0.147699 |
| review_scores_value | 0.145960 |
| beds | 0.144976 |
| review_scores_cleanliness | 0.143778 |
| review_scores_rating | 0.134326 |
Note: this table only shows the top ten features
Linear Models
- LinearRegression
- Ridge
- Bayesian Ridge Regression
Support Vector Machines
- SVR - Performed well with sparse data (hot-encoded)
Trees
- DecisionTreeRegressor
Emssembles
- RandomForestRegressor
- GradientBoostingRegressor
- HistGradientBoostingRegressor
- XGBRegressor
- LGBMRegressor
ANN
- Three hidden layers Neural Network
- r2
- mae
- mape
- mse
- rmse
| model | linear_regression | lasso | ridge | svr | decision_tree | random_forrest | gradient_boosting | hist_gradient_boosting | xgb | LGBM | ann_regressor |
|---|---|---|---|---|---|---|---|---|---|---|---|
| train_r2 | 0.5544 | 0.4966 | 0.5544 | 0.6094 | 0.6435 | 0.6564 | 0.896 | 0.7041 | 0.9565 | 0.8016 | 0.9491 |
| val_r2 | 0.5568 | 0.498 | 0.5568 | 0.6002 | 0.5877 | 0.6279 | 0.78 | 0.6824 | 0.803 | 0.7422 | 0.7845 |
| test_r2 | 0.5603 | 0.5012 | 0.5603 | 0.6077 | 0.5965 | 0.6362 | 0.7863 | 0.694 | 0.8111 | 0.7533 | 0.7942 |
| mean_yhat_val | 155.944 | 156.1785 | 155.9431 | 142.8075 | 155.889 | 155.8143 | 155.5398 | 155.985 | 155.6422 | 155.786 | 145.6884 |
| mean_yhat_test | 155.7705 | 155.7204 | 155.7696 | 142.5344 | 155.4132 | 155.9366 | 155.724 | 156.0552 | 155.6028 | 155.7839 | 145.4348 |
| val_mae | 0.3843 | 0.3997 | 0.3843 | 0.2811 | 0.3434 | 0.3332 | 0.2425 | 0.306 | 0.2263 | 0.2728 | 0.2141 |
| test_mae | 0.3817 | 0.3939 | 0.3817 | 0.2811 | 0.3419 | 0.332 | 0.2418 | 0.3029 | 0.225 | 0.2698 | 0.2118 |
| val_mse | 5427.405 | 6147.953 | 5427.3707 | 4895.9498 | 5049.7012 | 4557.1913 | 2694.7748 | 3889.0085 | 2412.8689 | 3156.8245 | 2639.4353 |
| test_mse | 5252.9704 | 5959.0155 | 5252.9621 | 4687.2693 | 4820.9274 | 4346.8873 | 2553.1095 | 3656.1552 | 2256.5164 | 2947.1248 | 2459.0037 |
| val_rmse | 73.6709 | 78.4089 | 73.6707 | 69.9711 | 71.0612 | 67.507 | 51.9112 | 62.3619 | 49.121 | 56.1856 | 51.3754 |
| test_rmse | 72.4774 | 77.1947 | 72.4773 | 68.4636 | 69.4329 | 65.9309 | 50.5283 | 60.4661 | 47.5028 | 54.2874 | 49.5883 |
| Model | mean_yhat_val | mean_yhat_test | train_r2 | val_r2 | test_r2 | val_mae | test_mae | val_mape | test_mape | val_mse | test_mse | val_rmse | test_rmse |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| linear_regression | 155.4831 | 156.4836 | 10.0 | 10.0 | 10.5 | 11.0 | 11.0 | 9.5 | 9.5 | 9.0 | 11.0 | 9.0 | 11.0 |
| ridge | 155.4825 | 156.483 | 10.0 | 10.0 | 10.5 | 10.0 | 10.0 | 9.5 | 9.5 | 10.0 | 10.0 | 10.0 | 10.0 |
| bayesian_ridge | 155.4768 | 156.4773 | 10.0 | 10.0 | 9.0 | 9.0 | 9.0 | 8.0 | 8.0 | 11.0 | 9.0 | 11.0 | 9.0 |
| svr | 168.6909 | 169.1026 | 5.0 | 6.0 | 6.0 | 8.0 | 8.0 | 11.0 | 11.0 | 6.0 | 6.0 | 6.0 | 6.0 |
| decision_tree | 154.8135 | 156.4564 | 8.0 | 8.0 | 8.0 | 7.0 | 7.0 | 7.0 | 7.0 | 8.0 | 8.0 | 8.0 | 8.0 |
| random_forest | 155.1075 | 156.3841 | 7.0 | 7.0 | 7.0 | 6.0 | 6.0 | 6.0 | 6.0 | 7.0 | 7.0 | 7.0 | 7.0 |
| gradient_boosting | 155.1426 | 156.318 | 3.0 | 2.0 | 3.0 | 3.0 | 3.0 | 3.0 | 3.0 | 2.0 | 3.0 | 2.0 | 3.0 |
| hist_gradient_boosting | 155.2242 | 156.4411 | 6.0 | 5.0 | 5.0 | 5.0 | 5.0 | 5.0 | 5.0 | 5.0 | 5.0 | 5.0 | 5.0 |
| xgb | 155.4133 | 156.6328 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 |
| LGBM | 155.0745 | 156.3721 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 | 4.0 |
| ann_regressor | 153.1353 | 154.2809 | 2.0 | 3.0 | 2.0 | 2.0 | 2.0 | 2.0 | 2.0 | 3.0 | 2.0 | 3.0 | 2.0 |
Best Hyperparameters:
- learning_rate: 0.05
- max_depth: 10
- n_estimators: 700
Metrics
- RMSE: 46.73
- r2: 0.96
- mean price: 155.65
- longitude is the feature mean highest contribution to the price
- private bathroom
- accommodates
- latitude
- bedrooms
- accommodates
- private bathroom
- longitude
- latitude
- bedrooms
Features contribution towards the prices - SHAP value
Features absolute contribution towards the prices sorted by the maximum absolute value SHAP value
Waterfall example of features contribution to a listing price
Features clustering mean SHAP value