# Predicting Airbnb Rental Prices Using Multiple Feature Modalities

**Authors:** Aditya Ahuja, Aditya Lahiri, Aniruddha Das  
**arXiv:** [2112.06430](https://arxiv.org/abs/2112.06430)  
**Date:** 2021-12-13  
**Summary page:** https://www.summarizepaper.com/en/paper/2112.06430v1/

## Key points

- Predicting the price of Airbnb rentals is crucial for hosts and customers
- Price prediction regression task has downstream applications, including recommending similar rentals based on price
- Proposed algorithm utilizes geolocation, temporal, visual, and natural language features
- Data collected from official Airbnb public dataset of hotel chains in eight cities across California
- Dataset includes host details, rental listing features, rental prices, and reviews
- Exploratory data analysis shows Los Angeles has the highest number of rentals followed by San Diego and San Francisco
- Most rentals can accommodate two people according to the accommodates feature
- Further analysis conducted to refine the model and handle outliers effectively in terms of prices
- Study aims to develop an accurate price prediction algorithm for Airbnb rentals using multiple feature modalities

## Summary

The task of predicting the price of a listed Airbnb rental is crucial for both hosts and customers. It allows hosts to set a reasonable price without compromising their profits, while also helping customers understand the key factors influencing the price and providing them with similarly priced options. This price prediction regression task has various downstream applications, including recommending similar rentals based on price. To create a reliable and accurate price prediction algorithm, this study proposes utilizing geolocation, temporal, visual, and natural language features. To gather data for this project, the researchers collected information from the official Airbnb public dataset of hotel chains in eight cities across California: San Diego, Los Angeles, San Mateo, San Francisco, Santa Monica, Santa Cruz, Pacific Grove and Oakland. The dataset included details about the host as well as features associated with each rental listing and the rental prices. Additionally reviews corresponding to each rental were incorporated into the dataset to enrich the feature space. In total the cumulative dataset consisted of 57000 listings across California and over 200000 reviews. Exploratory data analysis was conducted on the collected data. The distribution of listings among different cities was observed through a bar plot which showed that Los Angeles had the highest number of rentals followed by San Diego and San Francisco. The accommodates feature provided information about how many people an Airbnb rental can accommodate which was further analyzed through a histogram showing that most rentals could accommodate two people. In order to refine their model and handle outliers effectively in terms of prices further analysis was conducted. Overall this study aims to develop an accurate price prediction algorithm for Airbnb rentals by leveraging various features such as geolocation temporal factors visual aspects and natural language processing. By considering these multiple feature modalities the researchers hope to provide valuable insights for both hosts and customers in determining appropriate pricing strategies and facilitating informed decision-making in renting accommodations.