Airbnb, What Sells?

In 2008, Airbnb came onto the scene and allowed people to rent out their private houses and apartments for short-term stay. Given the vast number of different property types, is it possible to predict what the pricing for a listing might be? Can we use this data to figure out what consumers are looking for?
In order to help us figure out the answers to these questions, we’ll be using data supplied by Kaggle and Airbnb for the Boston area. In order to create a model to predict pricing, linear regression was employed. We came back with an R2 value of .70 when predicting price. This means our model fits the data moderately well, but what else can we learn from this?
One of the things we can perhaps learn from this is what the main features are that contribute the most to the price? Let’s start by looking at how correlated our data is to the price.

As we can see from these snippets above, a few of these are positively correlated and a few are negatively correlated to various degrees. ‘Accommodates’, that is, how many people the listing accommodates, number of bedrooms, beds, and room type all seem to be good candidates for main factors for predicting price. To be sure however, we can try removing these and all other features from our models one by one and see how our R2 changes.

Although all the changes to our R2 aren’t shown above, these are the features that showed the most significant impact to the R2 score. As we can see, some of the items strongly correlated to price also seem to affect our ability to predict price accurately, albeit individually not that much. That said, the number of bedrooms and bathrooms seem to be the top two features that contribute to the predictability of price. Let’s find out how many bedrooms and bathrooms people are typically looking for.

By a very large margin, 35,411 total reviews versus 6,280 for the next highest combination, it seems like many people look for a 1 bedroom, 1 bathroom combination when compared to all the other possible combinations these listings can possibly provide when com.

If you’re looking into listing your own property, these factors can help you determine how to price your listing and give you an idea of what people are looking for. Though there is more work to be done to get the full picture, I hope you gained some insights.

To further dive into how this analysis was done and perhaps customize and change the model to fit your own needs, please go to the project repository here.