Boston Airbnb Listing 2021

Table of Contents

• Project Background
• Author
• Data Source
• Libraries Used
• Motivation for The Project
• Files
• Summary of the results
• Acknowledgment

---

Project Background

This project is to analyze Airbnb listing data on Boston 2021. (Latest scraped data from Airbnb OpenData at the time of writing)

Oct 2021

For full article explaining this project, link below:

https://weimingchenzero.medium.com/inside-airbnb-boston-2021-building-a-machine-learning-price-predictor-model-for-boston-listings-d6bcbe13b0f9

---

Author

Weiming Chen (Alan)

---

Data Source

• http://insideairbnb.com/get-the-data.html
• http://data.insideairbnb.com/united-states/ma/boston/2021-09-19/data/listings.csv.gz (19 September, 2021)

---

Libraries Used

• numpy
• pandas
• matplotlib
• sklearn
• (Python 3.9 / Jupyter Notebook)

---

Motivation for The Project

The main goal is to build a Airbnb Listing Price Estimator model by implementing machine learning from Boston 2021 data to generate a LinearRegression model.

Host, agency, or end user can use this model to estimate / predict listing prices by inputing some independent variables. I also tried to find out what most standard amenities are listed in Boston listing that new host might considering for his/her new listing. And some others like: most demanded/popular districts; price ranges distribution.

---

Files

• /files/Boston_airbnb_2021/listings.csv
• boston_airbnb_2021_refactor.ipynb

---

Summary of the results

The Boston Airbnb Price Estimator/Predictor model I built based on this dataset is able to achieve 0.62 R-squared for training score, and 0.6 r-squared score for test score. The model can explain at least 60% of the variabilities of the data, could be a barebone working model in real life.

However, for more accountable on predictability (aiming for >0.7), I might need to train on more Boston past data and experiment different features selection combination as well as other cleaning strategies. (The model used 3000 instances/rows after cleaning and one-hot encoded categorical variables). I might aim for 6000 working(cleaned) instances to train.

Price range distribution on this dataset reveals that $60-80 most offering range; Overall, $80 - 160 range are popular; Higher standard $180-220 also quite common in Boston also; Listing offerings drop significantly after > $220. Listing price median is $141 which makes more sense than mean: $191 as it might be affected/stretched by outlier in the data ($10000.00 per night)

---

Acknowledgment

• Code snippet used about the function of train test, coefficients from Udacity example.
• Plot on price range distribution inspired by the book “Data Science from Scratch” - Visualization chapter.
• Some helper solution founds on StackOverflow.
• Airbnb OpenData