Predicting Student Attrition using Data Mining Predictive Models
This paper demonstrates how educational data mining can help institution in decision-making specifically to reduce student attrition. Phases of CRISP-DM (Cross Industry Standard Process for Data Mining) methodologyare followed in order to determine students at-risk of dropping out after the first semester in their freshmen year. Predictive models namely, decision tree, naïve bayes, and rule induction were built and applied to process the data set. Subsequently, these models were tested for accuracy using 10-fold cross validation. Results show that, given sufficient data and appropriate variables, these models are capable of predicting freshmen attrition with roughly 80% accuracy. Moreover, the average grades of the students can be used as predictor in determining student attrition unlike the gender attribute that yielded no significant result.
Keywords - Student attrition, Cross industry standard process-data mining, Decision tree, Rule induction, Naïve bayes,