Airbnb_project

Descriptive analysis of Airbnb data from Seattle, Boston and Copenhagen (see also @dima806/digging-into-airbnb-data-reviews-sentiments-superhosts-and-prices-prediction-part1-6c80ccb26c6a">this Medium post for detailed description)

Used libraries:

• collections
• dateutil
• eli5
• geopy
• matplotlib
• nltk
• numpy
• pandas
• scipy
• seaborn
• sklearn
• skopt
• tqdm
• warnings

Libraries installation with Pypi

pip install eli5

pip install geopy

pip install matplotlib

pip install nltk

pip install numpy

pip install pandas

pip install scipy

pip install seaborn

pip install sklearn

pip install tqdm

Motivation for the project

The aim of the project is to analyze the latest Airbnb data publicly available for three different cities (Seattle, Boston and Copenhagen), to perform sentiment analysis of the reviews for their customers, to reveal the difference between Airbnb superhosts from ordinary hosts, and to understands main factors responsible for the prise of Airbnb apartments.

Files in the repository

• airbnb_final_analysis_v3.ipynb - jupyter notebook with all details about preprocessing and analysis
• README.md - this file

Summary of the results of the analysis

• Overwhelming majority (> 95%) of Airbnb reviews are either positive or neutral.
• For Copenhagen, the the technical messages fraction about apartment cancellation is about 3.5%, or 4-6 times larger than in Boston and Seattle.
• Also, Copenhagen has much smaller fraction of superhosts (10%) than Seattle (40%) or Boston (23%).
• For all these cities, superhosts tend to have larger total and monthly averaged number of reviews, review scores and yearly availability are larger for superhosts than for ordinary hosts. On the other hand, the number of minimum nights, host response time and the host listings counts are smaller for superhosts than for ordinary hosts. This may reflect the higher popularity of superhosts and their higher level of service, compared to ordinary hosts.
• Among the most important features for daily price predictions are the distance to the city center and the type of the room. However, there are also significant differences between largest influencing features between different cities. For example, feature hosting_listings_count is valuable for US cities (especially for Boston) but is negligible for Copenhagen.
• Based on model trained by the data from different cities, we are able to predict the prices for a given city with a decent R2 score close to 0.7.

Acknowledgements

None.