Notes

Some info about Airbnb Airbnb is an online portal that connects people who look for a home or a room for short periods with people who have an extra space to rent, generally private. The site was opened in October 2007 by Brian Chesky, Joe Gebbia and Nathan Blecharczyk. The idea comes from the fact that in 2007 Brian and Joe moved to San Francisco; at that time the Industrial Design Society of America was organizing the annual conference and the availability of rooms in the hotel was exhausted. They were not able to pay their rent, so offered part of their apartment to other travelers interested in the conference. In January 2009 the company started thanks to the incubator Y Combinator. The continued its rapid growth, and in November 2010 received 7.2 million dollars from Greylock Partners and Sequoia Capital, reaching the milestone of 700,000 nights booked, the 80% in the last 6 months of 2010. In February 2011, the nights booked arrived at one million and the turnover increased by 65% compared to the previous month. 25 May 2011, the actor and partner of A-Grade Investments, Ashton Kutcher, announced a major investment in the company and its role as strategic brand ambassador. The company is in continuous international expansion: buying Accoleo, a German ambassador. In July 2011, Airbnb receives an additional $112 million dollars and the valuation of $1 billion. Currently, Airbnb has reached a quote share higher than $20 billion dollars, far above the value of many of the world's leading hotel chains.

The Case: Which Users Will Probably Book a First Experience with Airbnb? The new Airbnb users can reserve a location in more than 34,000 cities and in more than 190 countries. Being able to accurately predict whether a new user will book the first travel experience, Airbnb can:

• Share more personalized content to their community, aiming the communication on higher potential users.
• Reduce the average time to the first booking.
• Better predict the demand. For the analysis, a sample of users has been extracted with their demographics data, some web sessions records, and some summary statistics on users. The objective is indeed to predict how likely will be made a reservation by a new user. All users included in this analysis are in the US.

Available Data There are 12 possible outcomes in the data to generate the analysis target variable: the country of destination may in fact be "United States", "FR", "CA", ..., "NDF" (No destination) and "Other". "NDF" is different from "Others", because "Others" means that there was actually a reservation, but in a country not included in the list, while "NDF" means that there was no reservation. The training dataset contains a sample of 213,451 users with accounts created from 1/1/2010 until 30/6/2014. In the dataset containing the web sessions, the 01/01/2014 behavior data, while the users dataset dates back to 2010.

Reading of Data Files The analysis flow start from reading two important files: the data of users included in the analysis and the file relating to activity sessions recorded by users themselves from 1/1/2014 to 06/30/2014. The first analysis involves the verification of the correct data read and data quality for any subsequent data cleaning stages. In KNIME the Statistics node allows you to calculate the distributions of qualitative variables and some summary measures of quantitative variables in this case the unique numerical are ages in train users and secs_elapsed in sessions). The Interactive Tables node enables you to display the data files to better understand the nature of any problems in the data. The analysis shows the necessity of careful data preparation, working both on data cleansing and on the transformation of the available variables.

Processing of Relevant Data The first part of the analysis is linked to the need of selecting only active customers starting from 2014, and consider the sessions preceding their reservation. Before doing this, it has been necessary to obtain the date of the first activity business from the timestamp available, through appropriate string functions and subsequent transformation into a date field. Secondly, it has been calculated the number of seconds elapsed between the date of initial activity and the reservation date (with a daily level approximation); this figure will serve to isolate, at least by approximation, previous sessions with respect to the booking day. As can be seen from the data, about 1 out of 6 customers that book, book on the first day of activities on the website.

Selection of Active Visitors since 2014 The first part of the analysis, however, is linked to the necessity of selecting only active visitors since 2014, to select then the sessions before their reservation (considering the sessions dates available). From 213,451 visitors contained in the training set, 76,430 started surfing the website in 2014 and then they are selected for analysis. Using the number of seconds that have passed since their first action on the website (file sessions.csv) and the difference between the date of first visit previously built and the eventual date of reservation, we have selected the relevant sessions as predictors of booking. For visitors without booking all actions carried out in the six months available were considered.