Airbnb is an online platform where people can offer their apartments or houses to other people for short-term stay. In this setup, the company Airbnb connects both parties, the provider of the accomodation and the seeker and after a successful rent Airbnb gets a fee for each booking. The advantages of this way of staying in another city is among others to easily get in contact with locals through the landlord and to gain a better experience how locals actually live. For both the provider and the seeker of the accomodation it is important to get a better understanding what the real price of the property is. This article provides an in-depth analysis of the evolution of Airbnb in Munich and we present a model which can predict the price of a given accomodation. More technical details about this analysis and the model can be found at Github.
This analysis will be mainly from Airbnb which was gathered between 2009 to 2020. The original dataset has 11128 entries with 106 different features and can be downloaded from Kaggle. Features are the different informations about the property to rent such as for example the price, the location, or the average rating by others.
The analysis attempts to provide answers to the following questions:
- Did the AirbnB business grow and evolve in Munich?
- What are the most popular neighboorhoods in Munich and what makes them popular?
- Can we predict a listing price and what are the most important features to estimate Airbnb rental price?
Part 1: Did the AirbnB business grow and evolve in Munich?
In order to answer this question we first had a look at the total number of new listings at AirBnb in Munich since 2009 as depicted in Fig. 1. We can see that from 2009 till 2015 the number of new listings has steadily increased. 2016 and 2017 were years of decline and from then on the number of new listings was almost constant at 1000. In 2020 the number of new listings has descreased drastically which can most likely be explained by the Covid-19 pandemic.
In Fig. 2 we can see the average monthly price of an appartment in Munich between May 2020 and May 2021. We can see that the price has had a peak in September. Since the data we inspect was from May 2020 and the cancellation of the Oktoberfest for 2020 has not been decided yet we assume that this increased price can be explained by the Oktoberfest which actually takes place end of September. In total there was a steady increase in average from about 109€ to 118€ during the inspected period of time.
Part 2: What are the most popular neighboorhoods in Munich and what makes them popular?
In Fig.3–6 we have illustrated the Top 10 neighbourhoods by different measures: total count of listings, average listings price, review score location, and review score rating. For the review score location and review score rating the differences are not as clear as for total count of listings and average listing price.
The top neighborhood with the most listing is Ludwigsvorstadt-Isarvorstadt in Fig. 3, which is really central neighborhood and the central station is part of this neighbourhood, too. All remaining neighboorhoods in the Top 10 are allso quite close to the city center.
In Fig .4 the listing with the highest average price can be found in Altstadt-Lehel, which is the most central part of Munich, but it is also a rather small one. One outlier can be found in the Top 3: Trudering-Riem is far out from the city center. The high average price can be explained by its proximity to the exhibition center in Munich and the demand of accomodations close the an exhibition center.
The highest review score location is given for Altstadt-Lehel as depicted in Fig. 5 which is not really surprising since Altstadt-Lehel is the actual city center where most of the top spots such as the Frauenkirche and the Englischer Garten are located. The highest average review score rating is given for Neuhausen-Nymphenburg as you can see in Fig. 6.
As a concluding remark we want to highlight that Au-Haidhausen and Sendling are the only two neighbourhoods that appear in all four Top 10 lists but none of them is a top neighboorhood in the list.
Part 3: Can we predict a listing price and what are the most important features to estimate Airbnb rental price?
In order to estimate the price of an Airbnb rental in Munich, we have first cleaned the data from the listings by removing outliers, by dropping all the features with much missing data, and by filling the remaining unknown data with the mean or median value.
By doing so, we have fitted a model with a RandomForestRegressor. We optimized the parameters of the regressor with GridSearchCV. With that, we have created a model that has an error score of 173.
The list of the most important features for predicting the price can be found in Fig. 7. We can see that having a guest profile picture and requiring a guest phone verification has a big influence on the price. Having an airbed as well as the cancellation strategy also has a big impact on the price.
The error score still has some room for improvements such as implementing a better and more advanced prediction pipeline with different regressors.
In this blog we have analyzed the airbnb data from Munich, and shortly came the the following conclusions:
The number of new AirBnb hosts is increasing, but the rate of growth is stagnating. The closer the neighbourhoods to the city center, the more popular it is. We can predict the price of an accomodation with a RandomForestRegressor and found out that guest profile picture and guest phone verification both have a high influence on the price in Munich.
For me as a Munich local this analysis was really interesting, so I can just recommend you to do the same: go ahead and start to analyze the data of your hometown!