Discovering patterns from High School Academic Data: A Descriptive Data Mining Approach

Abstract

Although the educational data repository of Ethiopian High schools is very large, successfully implementing Data Mining (DM) still relatively new, which intended for identification, extraction of new and potentially valuable knowledge which helps to reduce student failure rates in high school subjects. This study apply descriptive DM and hybrid DM process model used to identifying, valid cluster analysis of students based on their academic performance in each subject they took. Association rule mining is also used discover interesting patterns and relationships between attributes/subjects and student gender which directs the decision making in education sector. Firstly, cluster analysis was performed to group the student into clusters based on its similarities. A total of four cluster analysis experiment was conducted on 5181 student academic record and 12 attribute used to build a model. The cluster result from these four experimentations are interpreted and evaluated. Among the four models, the one with K = 3, Seed value = 100 and Euclidean distance function has shown better segmentation of the students. This result of the model was selected according to objective measure SSE, number of iteration and subjective measure of expert judgments because the resulting student cluster give more insight to tackle quality of education problem. Secondly, association rules mining were implemented to different dataset to discover strong, interpretable rules that shows patterns of students result in each subjects. The PredictiveApriori algorithm generate rules with high predictive accuracy using RapidMiner Studio (RMS) and Apriori algorithm of WEKA results simple, interpretable and actionable rules shows strong correlation of antecedents and consequents subjects which achieves above the minimum support, confidence metrics and lift constraints. This new clusters pattern of students and association rules helps to give attention to similar student clusters, plan in advance related subject tutor and also improvising the subject score in each particular subject, in general improve quality of education and help policy makers as inputs.