The examples that were introduced this week are very inspiring - not only the actual application, but also the general approach and concerns. I would advice everybody to watch the whole conversation between George Simons and Vernon Smith at Educause 2011, not only the part by Vernon. George gives some very good concerns at the end of his talk. Just to mention one: "Analytics must be rooted in learning sciences".
I liked the Rio Pace example very much because they were able to create effective predictions using straightforward data and only a naive bayesian classifier. This is quite opposite to practice in educational research: Naive bayes is not a solid model, statistically, but it works fine in practice.
The Signals project at Purdue is quite interesting. We apply something similar in our portfolio system (called epass) for some of the schools, but we are not specificaly looking at retention rates.
This week I have played around with the free datamining tool Rapidminer using some of the usage data of our epass system for medical trainees (1.8 million data points over two years). I discovered a very clear picture of how the portfolio is used during the day: lunch break, dinner, small peak after dinner... But also that users spent more time using the portfolio when they log on in the weekend. It is only a start, but it is a nice way of getting into the tool.
Hi Jeroen, I fully agree with "analytics must be rooted in learning science". That is, why I was suprised, that Smith talks about Roles For Successful Predictive Modeling (slide 155) but does not list individual students and teachers, or education scientists among the stakeholders. Kathrin (lak12 participant)
ReplyDelete