Data Mining, Predictive Modeling, and Recruiting Targets

Le résumé fourni par la source

In an effort to make the best use of its financial and personnel resources, the Division of Graduate Studies at the University of Central Florida has started to take a new approach to better focus its recruiting dollars and strategies. While communicating with all admitted students at ucf, the Office of Graduate Recruiting, within the Division of Graduate Studies, wanted to target its recruiting dollars specifically on communications to those students who had the potential to be persuaded into attending ucf for their graduate education. A partnership was formed with the Department of Statistics and Actuarial Science to apply statistical methods to historical data to build a model that would allow the Division of Graduate Studies to predict which admitted students are likely to enroll in a ucf graduate program. A logistic regression model will be implemented to predict student enrollment for the fall 2007 semester. Targeted communications will also be sent to those students predicted to be undecided about attending ucf for graduate work. Communication interaction, survey results, and yield rates will be compared with current numbers to determine if this model and targeted communications are responsible for increases in enrollment. Knowing that of the students accepted each term into the University of Central Florida, some fraction would enroll, some might enroll, and others would choose not to enroll, the Division of Graduate Studies decided that if it had a model that predicted whether or not a particular student is likely to enroll, it could better allocate its resources to improve the overall quality of students who do enroll. For example, a fellowship may be offered to a particular high quality student who might otherwise be predicted not to enroll. After creative discussions and brainstorming sessions, a partnership was formed with the Department of Statistics and Actuarial Science. The initial goal was to apply statistical methods to historical data and then build a model that would allow the Division of Graduate Studies to predict which admitted students are likely to enroll in a ucf graduate program. This model would be used to improve student quality, more accurately predict enrollment totals and financial needs, and also allow for targeted recruiting dollars and communications. Because one of the enrollment goals of the Division of Graduate Studies is to increase enrollment in the science, technology, engineering, and math (stem) areas, the data analyzed in this study were specific to the stem graduate programs at ucf. Method The statistical method discussed here is a supervised learning technique, meaning it must be presented with a large number of observations (applications, in this case), each consisting of predictor variables and the response (whether or not the student enrolled). A mathematical model is then used to create a rule that can predict the response given only the predictor variables. The data used in this analysis were collected from several sources to form a single datamart containing all of the information deemed useful. This datamart consisted of 71,692 applications to UCF Graduate School programs from 1999 to 2005. Due to a significant change in the application procedures just before 2004, only 2004 and 2005 data are used to build a predictive model. The statistical software packages Enterprise Miner and SAS (SAS Institute Inc., Cary, NC), and R (R Foundation for Statistical Computing, Vienna, Austria) are used to analyze the data. The predictors include the following: specific graduate program and college to which student applied, academic level of program (e.g., doctoral, master's), gender, ethnic group, birthday, citizenship status, native country, whether student previously inquired about a ucf graduate program, Graduate Management Admission Test (gmat) score, Graduate Record Examinations (gre) score, Test of English as a Foreign Language (toefl) score, grade point average (gpa) for last 60 hours of undergraduate classes, numbers of colleges and programs to which student applied in current term and in previous terms, whether student went to a top 100 university as well as the university name and type of degree earned, whether student enrolled before at ucf, whether student completed undergraduate studies at ucf, and year and specific term for which student applied. …

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.