Predicting academic performance of undergraduate students in blended learning

Abstract—In this paper, we propose to predict the academic performance of university students from multiple-source data in multimodal and mixed learning environments using data fusion and extraction. We have collected data from 135 university students and different variables from four different sources. First, we applied data fusion and preprocessing to create a set of summary data in numerical and categorical format. We then applied different white box classification algorithms provided by Weka’s data mining tool to select the best algorithm. The PART algorithm shows the best performance of the observed quality metrics, (ROC Area =0.917). for the next test we applied attribute selection algorithms being ClassifierSubsetEval with J48 the best with a value of (ROC Area =0.9380), then we used two automatic learning algorithms Voting and Stacking given being the best voting result with the Jrip algorithm (ROC Area =0.9330) Finally, we show the best prediction model that is a hybrid between the classification and automatic learning algorithms with a value (ROC Area =0.9420) to help the instructor to take corrective actions with the students at risk of abandonment or failure.