Sas jmp customer churn answers
During the presentation, Bart displayed one time-lapsed section of the graph, and you could see side-by-side nodes turning red one after the other.īart Baesens, PhD, presents at Analytics 2011 When Bart's team first completed the network map and started looking at random samples of the data, they saw chains of churners popping up together right away. Are customers influenced only by direct neighbors in the network, or are they also influenced by friends of friends?.Were there social network effects in the data?.Some of the questions they wanted to answer were: (Text messages were considered a constant 5 seconds each.)
![sas jmp customer churn answers sas jmp customer churn answers](https://www.mdpi.com/futureinternet/futureinternet-13-00175/article_deploy/html/images/futureinternet-13-00175-g007.png)
The team created a network graph that showed weighted relationships between customers based on the number of seconds they talked to each other per month. Due to the complexity of determining relationships between so many entities, Bart says the analysis requires parallel computing and high performance analytics for the sparse matrix handling. The social network created based on this data included 8,000,000 edges, and the size of the data set was about 300 gigabytes in size. The data set Bart's team began working with included five months of call detail records on 2 million customers with a current churn rate of. "If we can find a way to bring social network data into simple logistic regression models, we could have a white box model that improves results," hypothesized the team.
![sas jmp customer churn answers sas jmp customer churn answers](https://www.ncsi.com/wp-content/uploads/2021/01/jmp15-power-of-jmp-pro_Page_1-232x300.jpg)
(At SAS we currently have 9 telecom customers using our Customer Link Analytics product to solve this business problem.) Enhancing models with network data The edges or ties can also be weighted to show the strength of the connection.
#Sas jmp customer churn answers series#
Social networks are typically illustrated through a series of nodes and edges (or ties) that show which people (or households or companies) are connected. Rather, it is a data mining technique that explores the patterns between people (or other entities) in a network or group. Specifically, they decided to enhance the existing regression models with social network data by asking questions like, who is my customer talking to? Who are his friends? Who is influencing him?īefore we go much further, I should mention that social network analysis does not typically analyze data from social network sites like Facebook and Twitter, as many people assume.
![sas jmp customer churn answers sas jmp customer churn answers](https://www.coursehero.com/thumb/92/f3/92f3c718df8a83f17a490513e3793d3beaaed993_180.jpg)
The performance of these models typically ranged from 70-90 percent accuracy with models that had ten to fifteen attributes.īart and his team then asked, "How can you improve those models?" One, you can improve your technique with complex modeling techniques or two, you can enrich your data. He found that many of the techniques currently in use perform quite similarly.
![sas jmp customer churn answers sas jmp customer churn answers](https://support.sas.com/training/us/paths/images/jmp-ls.jpg)
Wouter Verbeke, one of Bart's research students began the project by looking at existing local models for churn prediction. He presented the team's findings today at the Analytics 2011 conference in Orlando. Rather, customers are influenced by friends, friends of friends, and others within their network.īart Baesens, PhD, an Associate Professor at Katholieke Universiteit Leuven has been conducting research - along with a few of his PhD research students - to determine whether network analysis can improve traditional churn prediction models in the telecommunications industry. But individual customers are not isolated entities. Traditional churn models - designed to predict whether or not customers will cancel your company's services - treat customers as isolated entities.