To whom we predict
Quite often we have come across with "prediction models", which are able to describe events dating years, even decades back. We know precisely the number of students, or year of graduation, how long it took to complete the studies, what was the average grade or the drop-out rate for that year. When there is a sufficiently large amount of such data, we generate averages that we believe can help predict future student progress, lead degree programs, or support study guidance. In reality, however, we have lost the individual, the progress of which we should look at and, at the same time, have forgotten that a real prediction model should deal with new individuals starting their studies.
Often, learning prediction models are targeted as tools for study guidance, curriculum management, or as a university-level funding model tool. Surprisingly, the support provided for the customer or individual student's forecasting model may be forgotten and even more often the use of the forecasting model from the end-user perspective will be forgotten: the forecasting model should be able to produce reliable information for the development of educational products so that the industry in need of graduating students has access to experts with a suitable and up-to-date knowledge profile. As we currently anticipate the progress of studies, we may be focusing too much on the credits produced, the number of graduates, the duration of the studies, or the cost of teaching. A good prediction model would be able to highlight the changes needed in the knowledge profile in addition to these and would also consider the end user’s perspective. A new challenge is the increasing diversity of distance-learning programs, where students progress in digital environments according to their own schedules and whose content can be updated in real time anywhere in the calendar year. Regardless of the learning environment, the prediction model should be able to take into account that someone needs these graduating students!
Does the system support our operations or does it control what we do
A very classic question when developing information systems is how much flexibility should the system have to suit the needs of different users? From a purely analytical and statistical point of view, a strictly rule-based and locked system would be the easiest, but there are at least two big variables in the prediction model that requires system flexibility. First of all, the student’s background and life situation make an individual who, in any case, follows his or her own paths. Secondly, the degree structures underlying the forecasting model and the courses included therein will at least hopefully evolve to meet the needs of the end user mentioned above. Therefore, the system behind the forecasting model must allow and recognize a wide range of flexibility regarding individual choices, changes and updates in degree programme structures, study modules with variable scopes, and study paths originally designed to be of varying lengths. It would be unsatisfying if the system were so rigid that students would be forced into a particular template or that flexibility in their studies would be "system controlled". Undoubtedly, allowing flexibility makes it difficult to build prediction model algorithms, but only by allowing flexibility can one produce forecast information that serves the entire chain from the customer to the end user.
In order to meet the needs described, we have defined, for example, in the description of the AnalyticsAI project, the following: “The final phase of the project will provide a set of non-organizational and generic definitions of common ERP-critical information content and an e-PSP (electronic Personal Study Plan) prediction model”.
Let’s keep this in mind as we move forward with the project together.
Harri Eskelinen & Terho Lassila
Lappeenranta-Lahti University of Technology LUT