Building an integrated end-to-end ML pipeline for customer support optimization

Hi everyone, we are happy to announce our next meetup. As usual the meetup will be open for all to attend, and newcomers / beginners are very welcome.

The agenda for this meetup is:

18.30 Open Doors
18.50 Welcome / AOB
19.00 Sven's talk (+ questions along the way)
19.45 Break
20:00 Meet the Expert
20.30 Getting ready for 🍕 & 🍝 at https://via-toselli.de/

At this meetup Sven Thies will present a recently completed project from his work.

** Building an integrated end-to-end machine learning pipeline for customer support optimization

Training and testing of machine learning models is relatively easy, especially with R's well suited libraries, e.g. caret. However, if the end user of a machine learning model is not another machine but humans – especially those who don’t know much about machine learning models – our experience is that the main focus should not be on training and testing the model (e.g. pushing accuracy to the limit) but on how you can ensure that the user will use your predictions to create value for your company. Hence, this talk will be on how we leverage R in large parts of an integrated machine learning pipeline that utilizes a machine learning model for customer churn predictions to improve our customer support. In detail we will show how we

• leverage RStudio's Shiny to quickly build a web application that shows the customer support which customer is likely to churn (in the form of model predictions),
• use LIME (local interpretable model-agnostic explanations) to show why a customer is likely to churn and hence make the model predictions trustworthy to the customer support and
• implement a connection of the Shiny app to a backend database to constantly get feedback on the performance of the models predictions.

Further, we will quickly take a look on

• where we host RStudios Shiny for internal use and
• how we leverage Apache Airflow to keep the information in the Shiny app up to date.

About the speaker: Sven is a data scientist at Traum-Ferienwohnungen GmbH, a medium-sized company providing a marketplace for rental apartments in over 70 countries, which is located in Bremen. His background is in applied statistics and business administration. He is very passionate about working as a data scientist but equally on teaching and talking about this exciting field to share thoughts of utilizing data in organizations – e.g. building and integrating data science teams into the organizational structure. The tool which manly comes to action in his work is R and its ecosystem.