An Enhanced Multiple Classifier System For Predicting Students’ Academic Outcome

Abstract

Data-driven decision support approaches have been increasingly employed in recent years to unveil purposeful task-oriented patterns from accumulated students’ academic records. Educational data mining and Knowledge Discovery in Database (KDD) provide a viable solution to decipher implicit knowledge through predictive modelling. Most approaches used the induction of the individual classification models also known as the single classifiers and few efforts focuses on the ensemble methods; which are unstable, overfits and susceptible to skewed data or a poorly performing classifier. To overcome this work examines the existing multiple classification algorithms, Pandey and Taruna approach and proposes an enhanced MCS that uses bagging strategy to reduce variance and perhaps improves accuracy of the resulting model to improve the accuracy. The approach is based on the definition of a model that integrates several classification techniques to predict academic performance of undergraduate students. Data were collected, cleaned, preprocessed and integrated using MS (excel and access) and Weka software. Experiments were carried out using WEKA and EMCS improved using python library. The result shows that the proposed (EMCS) has an increase of 1.03% and 0.63% of accuracy over the Bagging MCS and Pandy& Taruna respectively