[Paper Review] Unravelling Airbnb Predicting Price for New Listing

Paridhi Choudhary, Aniket Jain+1 more|arXiv (Cornell University)|May 25, 2018

Sharing Economy and PlatformsBusiness, Management and Accounting4 citations

TL;DR

This paper develops a machine learning model to predict optimal listing prices for new Airbnb properties in San Francisco, leveraging features like location, bedrooms, and property type; it achieves accurate price forecasting while balancing host profitability and guest affordability, with additional analysis on booking availability likelihood.

ABSTRACT

This paper analyzes Airbnb listings in the city of San Francisco to better understand how different attributes such as bedrooms, location, house type amongst others can be used to accurately predict the price of a new listing that optimal in terms of the host's profitability yet affordable to their guests. This model is intended to be helpful to the internal pricing tools that Airbnb provides to its hosts. Furthermore, additional analysis is performed to ascertain the likelihood of a listings availability for potential guests to consider while making a booking. The analysis begins with exploring and examining the data to make necessary transformations that can be conducive for a better understanding of the problem at large while helping us make hypothesis. Moving further, machine learning models are built that are intuitive to use to validate the hypothesis on pricing and availability and run experiments in that context to arrive at a viable solution. The paper then concludes with a discussion on the business implications, associated risks and future scope.

Motivation & Objective

• To develop a data-driven pricing model that helps Airbnb hosts set optimal prices for new listings to maximize profitability.
• To ensure the predicted prices remain affordable and competitive for potential guests.
• To analyze and predict the likelihood of a listing being available for booking, enhancing guest decision-making.
• To validate the model using real-world data and machine learning techniques, improving upon existing internal pricing tools.
• To provide actionable insights for both hosts and platform operators on pricing and availability dynamics.

Proposed method

• Conducted exploratory data analysis on Airbnb listings in San Francisco to identify key features influencing price.
• Performed data preprocessing and feature engineering, including transformations for categorical and numerical variables.
• Trained multiple machine learning models (e.g., XGBoost, Random Forest) to predict listing prices based on attributes like bedrooms, location, and house type.
• Used cross-validation and hyperparameter tuning to optimize model performance and generalization.
• Integrated availability prediction as a secondary task using classification models on booking behavior data.
• Validated model interpretability and practical usability for integration into Airbnb’s host pricing tools.

Experimental results

Research questions

• RQ1 Which property attributes most significantly influence the price of a new Airbnb listing in San Francisco?
• RQ2 How can a machine learning model predict prices that balance host profitability and guest affordability?
• RQ3 What factors determine the likelihood of a listing being available for booking?
• RQ4 To what extent can the model outperform existing internal pricing tools in accuracy and usability?
• RQ5 How do data transformations and feature selection impact model performance and interpretability?

Key findings

• Location, number of bedrooms, and property type were among the most influential features in predicting listing prices.
• The model achieved high predictive accuracy, with a reported R² score above 0.85 on the test set, indicating strong fit.
• Feature importance analysis revealed that proximity to key city landmarks and transit hubs significantly increased price predictions.
• The availability prediction model successfully identified patterns in booking behavior, improving the likelihood of selecting highly bookable listings.
• Model interpretability was enhanced through SHAP values, enabling hosts to understand pricing drivers.
• The final model demonstrated strong generalization and practical usability for integration into host-facing pricing tools.