A school is having difficulty identifying, with a high degree of recall, to which students extra resources should be allocated to help prevent them from failing a course. In reality, it is difficult for a teacher or administrator to manually identify every student who is likely to need extra support given the large number of factors to consider when making such a decision with the added complexity of not necessarily knowing what those factors are.
A school may want to more completely identify students who are "at-risk" (likely to fail; final grade below 50%) and provide interventations (i.e. allocate resources) so that they stand a better chance of succeeding, while simultaneously controlling for false positives - allocating resources to students who would have likely passed irrespective of the interention is costly.
We'll create a predictive model based behaviour and demographics, and in the future, grading data available at the end of each of the first two terms. At this point, user experience is not important. It is important, however, to try and provide models that give probability-based estimates so that uncertainty in riskiness is effectively communicated. Methods to understand the appropriate tradeoff between precision (correctly identifying at-risk students) and recall (identifying as many at-risk students as possible) will be leveraged to help estimate and control cost.
It's not yet clear how the model will be operationalized but it would be advantageous to deploy to deploy models through a graphical or interactive tool.