youngjeong46 / airbnb-star-prediction

This classification model predicts whether an AirBnB listing in Seattle will receive a 5-star rating

Predicting Airbnb 5 Star Ratings

Overview

This project took Seattle AirBnB listings data and used it to build a classification model. The model predicts whether a listing for AirBnB will receive a 5-star ratings or not. Blog post on the project can be found here.

I also created a Web Application where users can input information about their (potential) listings and find out the probability of it receiving a 5-star ratings. The app will also suggest one amenity addition that will boost the probability the highest. You can see the web app github repo here, and see the application in action here.

Project Design

Having previously been an AirBnB host, I know the importance of obtaining and maintaining a 5-star Rating. It leads to an opportunity for the SuperHost designation, a status which leads to more search exposure and about 50% increase in monthly income. I believe how a host curates his or her listings and what they provide is reflected in the ratings and decided to put the theory to a test.

Insideairbnb.com is an independent research company focused on data visualization of AirBnB and its impact in major cities. They have collected a nice dataset on AirBnB listings in those cities, including Seattle. Using that data, along with Walkscore.com's API, I collected a dataset that I believe had features correlate with ratings. I ran various classification models with parameter tuning via Cross Validation. The models include:

• Logistic Regression
• K-Nearest Neighbors
• Support Vector Machine
• Decision Trees
• Random Forest
• Gradient Boosting
• Adaptive Boosting
• Catboosting
• XG Boosting

TThe metric I used to select the best model was the ROC-AUC score (Receiver Operating Characteristics Area Under the Curve). The best models using the ROC-AUC as the metric were the boosting CART models, with the XGBoost scoring 0.7891, followed by Catboost at 0.7837, Gradient Boosting at 0.7795, and Adaptive Boosting at 0.7757. Even though I could have chosen XGBoost as my model, the difference in metric performance wasn't enough to make a decision here. Therefore, I also chose to look at time performance (in terms of test predictions).

Comparing the time performance between the 4 Boosting Models, Gradient Boosting performed the fastest at 0.8168 seconds. It was 3 times faster than next fastest model (Adaptive boost at 2.39 seconds), and more than 8 times faster than XGBoost (6.84 seconds). Considering product implementation, I chose the Gradient Boost as the model of choice for its excellent timing performance.

Using the model produced I made predictions on the holdout (test) set. The ROC-AUC score for the test set was 0.71.

Some potential improvements can be made, and some that I want to implement include Description NLP to be able to suggest keywords for hosts to describe their listings with and Zillow/Redfin Integration for potential hosts that wants to purchase properties with the sole purpose of running AirBnB rental units.

Tools

• Data Collection: Google Geocode API, Walkscore.com API
• Data Storage: Pandas, pickle
• Data Visualization: Seaborn, Matplotlib
• Data Analysis: Scikit-learn, XGBoost, Catboost
• Presentation: Google Slides
• Web Application: Flask, Bootstrap

Project Organization

The structure of the project organization is adapted from Cliff Clive's DataScienceMVP, which itself is adapted from the famous Data Science project structure from Cookiecutters that you can access here.

    ├── LICENSE
    ├── README.md          <- The top-level README for developers using this project.
    ├── main.py            <- The top-level src code for this project.
    ├── data
    │   ├── interim        <- Intermediate data that has been transformed.
    │   ├── processed      <- The final, canonical data sets for modeling.
    │   └── raw            <- The original, immutable data dump.
    │
    ├── results            <- Trained and serialized models, model predictions, or model summaries
    │   ├── models         <- Generated model records stored in pickle files
    │   └── records        <- All generated records about the project in pickle files
    │
    ├── notebooks          <- Jupyter notebooks. Naming convention is a number (for ordering),
    │                         The step in the workflow (i.e. Obtain_Data)
    │                         `1_Obtain_Data.ipynb`.
    │
    ├── references         <- Data dictionaries, manuals, and all other explanatory materials.
    │
    ├── reports            <- Generated analysis as HTML, PDF, LaTeX, etc.
    │   ├── figures        <- Generated graphics and figures to be used in reporting
    │   ├── MVP.ipynb      <- The general overview of project in Jupyter Notebook.
    │   ├── slides.pdf     <- Slide deck for a short presentation.
    │   └── summary.pdf    <- summary of the project
    │
    ├── src                <- Source code for use in this project.

Appendix: Features Used

Features	Type	Description
neighborhood_group_cleansed	Str	Neighborhood in Seattle
room_type	Str	Type of room offered (Room or whole house)
bedrooms	Int	Number of bedrooms
bathrooms	Float	Number of bathrooms
beds	Int	Number of beds
price	Int	Nightly price of the listings (in Dollars)
security_deposit	Int	Security Deposit required (in Dollars)
cleaning_fee	Int	Cleaning Fee charged (in Dollars)
extra_people	Int	Charge for having extra people (in Dollars)
minimum_nights	Int	Minimum nights required for a reservation
calendar_updated	Int	how often the host updates the availability calendar
availability_30	Int	How many days are available for reservation in the next 30 Days
availability_90	Int	How many days are available for reservation in the next 90 Days
cancellation_policy	Bool	Is the cancellation policy flexible or not? (True if yes)
instant_bookable	Bool	Is the listing instantly bookable without host's review? (True if yes)
Bikescore	Int	A listing's Bikescore
Walkscore	Int	A listing's Walkscore
24-hour check-in	Bool	Does the listing provide A/C? (1 if yes, 0 if no)
BBQ grill	Bool	Does the listing provide a BBQ Grill? (1 if yes, 0 if no)
Bed linens	Bool	Does the listing provide extra Bed linens? (1 if yes, 0 if no)
Cable TV	Bool	Does the listing provide cable TV? (1 if yes, 0 if no)
Coffee Maker	Bool	Does the listing provide a coffee maker? (1 if yes, 0 if no)
Dishwasher	Bool	Does the listing provide a dishwasher? (1 if yes, 0 if no)
Elevator	Bool	Is there an elevator access for the listing? (1 if yes, 0 if no)
Extra pillows and blankets	Bool	Does the listing provide extra pillows/blankets? (1 if yes, 0 if no)
Family/kid friendly	Bool	Is the listing family and/or kid-friendly? (1 if yes, 0 if no)
Fire extinguisher	Bool	Does the listing provide a fire extinguisher? (1 if yes, 0 if no)
First aid kit	Bool	Does the listing provide a first aid kit? (1 if yes, 0 if no)
Gym	Bool	Does the listing provide gym access? (1 if yes, 0 if no)
Indoor fireplace	Bool	Does the listing provide indoor fireplace? (1 if yes, 0 if no)
Internet	Bool	Does the listing provide internet? (1 if yes, 0 if no)
Keypad	Bool	Does the listing provide a keypad entry access? (1 if yes, 0 if no)
Lock on bedroom door	Bool	Does the listing have locks on its bedroom(s)? (1 if yes, 0 if no)
Lockbox	Bool	Does the listing provide a lockbox for access? (1 if yes, 0 if no)
Long term stays allowed	Bool	Does the listing allow for long term stay? (1 if yes, 0 if no)
Luggage dropoff allowed	Bool	Is luggage dropoff allowed prior to reservation? (1 if yes, 0 if no)
Pack 'n play/travel crib	Bool	Does the listing provide Pack'n Play or travel crib? (1 if yes, 0 if no)
Patio or balcony	Bool	Does the listing have a patio or balcony? (1 if yes, 0 if no)
Pets	Bool	Does the listing allow pets? (1 if yes, 0 if no)
Private entrance	Bool	Does the listing have a private entrance? (1 if yes, 0 if no)
Private living room	Bool	Does the listing have a private living room? (1 if yes, 0 if no)
Refrigerator	Bool	Does the listing come with a refrigerator? (1 if yes, 0 if no)
Safety card	Bool	Does the listing provide a safety card? (1 if yes, 0 if no)
self check-in	Bool	Does the listing have self check-in? (1 if yes, 0 if no)