Authors:
Eitel J. M. Lauría
1
;
Eric Stenton
2
and
Edward Presutti
3
Affiliations:
1
School of Computer Science & Mathematics, Marist College, Poughkeepsie, NY, U.S.A.
;
2
School of Computer Science & Mathematics, Marist College, Poughkeepsie, NY, U.S.A., Data Science & Analytics Group, Marist College, Poughkeepsie, NY, U.S.A.
;
3
Data Science & Analytics Group, Marist College, Poughkeepsie, NY, U.S.A.
Keyword(s):
Early Detection, Student Retention, Freshmen Attrition, Predictive Modeling, Machine Learning.
Abstract:
We explore the use of a two-stage classification framework to improve predictions of freshmen attrition at the beginning of the Spring semester. The proposed framework builds a Fall semester classifier using machine learning algorithms and freshmen student data, and subsequently attempts to improve the predictions of Spring attrition by including as predictor of the Spring classifier an error measure resulting from the discrepancy between Fall predictions of attrition and actual attrition. The paper describes the proposed method and shows how to organize the data for training and testing and demonstrate how it can be used for prediction. Experimental tests are carried out using several classification algorithms, to explore the validity and potential of the approach and gauge the increase in predictive power it introduces.