Analysis of consumption patterns in Test-Achats website and Classification

Test-Achats is a Belgian consumer association founded in 1957 and one of the magazines published by this association. Quite known for the master-purchase which is the product or service that represents the best value in the magazine's comparative tests. Test-Achats employs more than 400 people and reports a total of more than 330000 subscribers through five magazines. Test-Achats also invests considerably in the development of its digital platforms. Since 2014 a tracking system has been set up on the website. Each action is recorded in the tracking file of the day. To these actions, various variables are associated like a device ID which makes it possible to identify a user who already came to the website. Through this thesis, these data are explored to better understand the various behavior of users. The main goal being to find how can we better convert prospects into members. First of all, it's important to understand data we have. We'll explain which data will we use and we'll quickly present data through the website. Then we'll analyze the different features related to the consumption of the website such as the time a user spent on the website. Clusters based on the section will be created, which will allow us to analyze the features for each these clusters. After these general analyzes, we will focus on the users journeys on the website meaning the temporal succession of articles they read. This will allow us to determine if users tend to stay in specific section of the site or if they are mobile on it and identify sections with strong links between them. Through this analysis, we will also try to understand the main differences between a members and a non members. In the second part, we will try to understand how can we better convert prospects into members. In order to do that, classification models will be used. We'll be interested if it's possible to separate the users who didn't register and those who became member. The data set being unbalanced different solutions will be proposed and many tests will be done to understand what is the best model. As this thesis is a concrete project, this report is deliberately oriented towards practice and a depth understanding of all manipulations performed.