Predicting academic performance: A bootstrapping approach for learning dynamic bayesian networks

Abstract

Predicting academic performance requires utilization of student related data and the accurate identification of the key issues regarding such data can enhance the prediction process. In this paper, we proposed a bootstrapped resampling approach for predicting the academic performance of university students using probabilistic modeling taking into consideration the bias issue of educational datasets. We include in this investigation students’ data at admission level, Year 1 and Year 2, respectively. For the purpose of modeling academic performance, we first address the imbalanced time series of educational datasets with a resampling method using bootstrap aggregating (bagging). We then ascertain the Bayesian network structure from the resampled dataset to compare the efficiency of our proposed approach with the original data approach. Hence, one interesting outcome was that of learning and testing the Bayesian model from the bootstrapped time series data. The prediction results were improved dramatically, especially for the minority class, which was for identifying the high risk of failing students.