Al-based churn prediction for training programs at VDAB
For VDAB, Belgium's job placement agency, delaware has developed an Al-based churn prediction model to experiment with. Based on historical data from participants and subscription data from the training itself, the model could help the agency to accurately estimate the churn rate of each training course.
"Predicting the likelihood that someone will prematurely leave the training doesn't really help all that much. You need to know why they are giving up as well." Says Charlotte De Vlieghere, data scientist at delaware. "To solve this problem, we added an explanatory model. Where the existing model only determines the risk of churn, the second model also identifies general churn drivers and determines how all these different factors impact the churn rate. These insights help to identify which actions are likely to have the most impact in any given situation.
"Interestingly, we discovered that only a handful of factors were enough to predict the majority of churn. In the nurse training curriculum, for example, access to daycare for participants' children proved to be decisive. This is due to the fact that nursing training lasts 3 years and thus requires a significant commitment. In the masonry curriculum, which is a lot shorter, it quickly became clear that most students who abandon the course early did not have any prior experience. A possible solution VDAB could look into is to implement the preliminary training as an extra option in the course."
Simon Claus, project leader at VDAB, looks back on a successful collaboration: "delaware managed to communicate the technical setup as well as the results of the predictive model in a very clear and concise way. As a result, we were able to translate these into concrete steps aimed at further improving our offering."