Capture Blog

How To Cheat at Enrollment Predictions, Part 3: Predicting Closer To the Target Date

This is the third installment of “How To Cheat at Predictions,” a look at four ways Capture Higher Ed’s competitors make their predictive models look better than they really are. As we have seen, the first way is to use leakers; and the second is to train and test a model with the same data.

Today, we highlight the third way — predicting closer to the target date.

In a recent Envision Challenge, Capture went up against a competitor and came up a little short. We don’t like coming up short, but we were ready to concede defeat … that is, until we realized what our competitor had done.

While we were asked to make an enrollment prediction immediately after admittance, the competitor was able to take advantage of the data gathered all the way up to enrollment.

That’s a lot easier to do. We were shooting at the target from 500 feet; they were shooting from 50.

Or since Capture is in the heart of college basketball land, a different analogy: We were predicting the winner of March Madness before the tournament. Our competitor was predicting the winner at the Final 4.

So if given the choice, always make your prediction as close to the target date as possible. Don’t be like us and expose yourself to the risk of making a prediction several months out.

We actually like the risk. With Envision, we have built the most advanced predictive engine in the higher education marketplace today, and we love any and every opportunity to prove it.

Take the Envision Challenge and let us run your data for free. See for yourself how we use models beyond logistic regression, multiple types of data, automatic feature selections, and dynamic scoring as well as how we can retrain your model within the enrollment cycle.

And be sure to tune in tomorrow for the series finale of “How To Cheat at Predictions” in enrollment management when we discuss the sneaky practice of “predicting enrollments using prospects.” It is another way our competitors get good accuracy statistics with models that crumble when trying to predict next year’s class.

By John Foster, Data Analyst, Capture Higher Ed

Share this!Share on FacebookTweet about this on TwitterShare on Google+Share on LinkedIn
Please wait...

Subscribe to our weekly blog digest!

You will get fresh weekly content, and we will never sell your email.