Machine Learning-based Categorization of Airbnb Listings in NYC

Abstract

This research focuses into creating a machine-learning-driven system to categorize Airbnb listings in New York City (NYC) based on neighborhood attributes and listing features. Utilizing data scraped from InsideAirbnb.com, including custom attributes such as median household income, craft beer and specialty coffee counts, and a connectivity score, KMeans clustering was applied to classify listings into four groups. These groups, named Normal People, The 2%, Central Action, and Hip Kids, offer insights into the city’s diverse landscape of Airbnb offerings. The classification model’s accuracy was validated using various semi-supervised learning techniques, resulting in 100% accuracy for some models. Dropping significant features like income in validation tests reduced accuracy to 66-78%, showing the importance of feature selection. The study demonstrates the potential of machine learning in enhancing Airbnb’s understanding of customer preferences and refining inventory management.