(This is Part 5 of a series looking at how enrollment predictions are putting higher education institutions at risk. Don’t miss our previous entries, the Series Introduction; Part 1, Treating Adolescent Decision-Making as Linear; Part 2, Not Adequately Testing Models in Real World Scenarios; Part 3, Trying to Forecast October’s Weather on January 1; and Part 4, Choosing Interpretability Over Accuracy.)
Bringing Too Few Tools to the Job Site
As we’ve discussed previously, the challenges presented by the non-linear relationships in enrollment management render a “linear-or-bust” toolset obsolete. The good news is that recent advances in computer science and statistics have produced a large set of predictive algorithms that can apprehend and utilize these non-linear relationships in data to make more accurate predictions than can be made when we assume relationships are linear.
But these algorithms all pick up on relationships in different ways, making some algorithms better suited for certain types of data. There is no one-size-fits-all master algorithm that always works best regardless of what we are predicting. So we must be smart with how we build our models to reflect this reality.
Rather than testing our data against one or two statistical approaches, Capture Higher Ed uses an approach that selects from dozens of algorithms in order to make the most accurate predictions using a partner’s data. This type of approach is made possible through advances in the field of machine learning, a subfield of computer science concerned with helping computers make increasingly optimal decisions using past information.
To put machine learning in the context of human learning, imagine when a child touches a hot stove. In an instant, she receives data (“ouch, that hurts”), considers what that data might mean (“stoves are hot and can hurt me”) and will likely make a better decision concerning stoves in the future.
Computers are capable of analyzing data in much the same way by looking at past data to predict future behaviors. You have surely seen examples of machine learning everywhere from your Netflix queue, to Google’s self-driving car, to Watson’s complete annihilation of Ken Jennings on Jeopardy.
At the core of machine learning is an emphasis on an empirical approach to prediction that focuses on deeply “listening” to the data to discern the signal from the noise, rather than a top-down, theoretical approach that creates a hypothesis to be tested from a set of “known” (or at least previously assumed) predictors.
The enrollment patterns of adolescents worldwide are complex and the most sophisticated tools should be utilized to accurately make predictions in this environment. With advances in computer science, statistics and computational processing, there is no reason why enrollment managers should limit their use of the most advanced technologies, which in the hands of a capable modeler, can produce highly accurate predictions.
By Thom Golden, Ph.D., Vice President of Data Science; Brad Weiner, Ph.D., Director of Data Science; and Pete Barwis, Ph.D., Senior Data Scientist, Capture Higher Ed