Improved Cluster Analysis for Graduation Prediction using Ensemble Approach

Predicting student performance has been one of major subjects in the educational data mining, for which a bucket of analytical methods has been proposed. Among these, a recent framework of bi-level learning is recently introduced with improved classification performance from a basic supervised paradigm. However, only k-means is exploited to derive data clusters, which are employed as references for context-specific classification modeling. As such, this paper presents an original work that applies ensemble clustering to deliver more accurate data partition, thus lifting the predictive accuracy. Based on data collected from Mae Fah Luang University databases, the new approach are usually more effective, especially to the minority class that is the core of imbalance problem. Besides, a parameter analysis is briefly addressed herein to specify recommended settings for future exploitation.