Consortium for Student Retention Data Exchange

1700 Asp Ave.
Norman, OK 73072

Phone: 405-325-2158
Fax: 405-325-7309
csrde@ou.edu

Predicting Academic Success Using CART: Classification and Regression Trees

presented by

Date of webinar: December 12, 2012

*Materials for this webinar are not currently available.

REGISTRATION INFORMATION
To register for a Webinar click here, download the form for the year you are registering for, fill it out, and either mail or fax it back to us.

Abstract

We apply a classification and regression algorithm in an effort to successfully predict if a student will or will not graduate. Specifically, the CART (Classification and Regression Trees) algorithm was applied to university retention data. CART is a nonparametric approach to classification and regression problems. It is a very robust method of performing classification. CART has several features that make it attractive over more traditional methods. Traditional methods of dealing with this problem often lack flexibility. Observations for example, are often assumed to be normally distributed. Traditional methods typically cannot deal with observations that contain categorical data or missing data in a natural way. CART analysis often makes progress on high dimensional data sets when other methods do not. The flexible nonparametric approach of CART will be discussed. The classification rules appear in the form of binary trees, which are easy to understand and interpret. Such decision tools may prove useful to colleges and universities trying to improve their retention challenges. CART techniques have been applied to several years of data from a small private undergraduate institution. Academic program specific models for predicting graduation indicate the potential to improve graduation rates by an average of twenty percent across the academic programs investigated. The CART regression algorithm will also be discussed as a useful tool for identifying students at risk.

The Consortium for Student Retention Data Exchange at the University of Oklahoma is a consortium of two-year and four-year institutions dedicated to achieving the highest levels of student success through collaboratively sharing data, knowledge, and innovation.

CSRDE is a sponsored activity of the Center for Institutional Data Exchange and Analysis at