Projecting student success: predictive analytic explorations of admissions data, the LASSI-HS, and institutional academic performance

There is an increasing need for student success professionals to have accurate data on students who are most likely at-risk in order to promote outreach efforts and interventions to those who may benefit the most. Using predictive analytics, university personnel can potentially identify these students on a large scale, allowing for targeted interventions, maximizing impact with limited institutional resources. The goal of the study was to attempt to build “real-time” models predicting for graduation that adapted as new information about a student cohort was available. Utilizing random forest algorithms, conditional inference trees, and logistic regression, this study explored the possibility of utilizing these techniques in a new institutional context. The first models started with low predictive accuracy, moving to moderate predictive accuracy as additional student data was incorporated. Variables related to high school performance and first-year collegiate performances were the strongest predictors found in the models. Survey data on student attitudes towards learning was not found to be a strong predictor overall, although it may be more predictive for students with lower high school GPA. Future studies should consider disaggregating the data further to see if certain variables predict with higher accuracy for specific student groups, such as students below a GPA threshold, or in a particular academic program.