A Machine Learning Based Framework for Predicting Student’s Academic Performance

Abstract

Educational Data Mining (EDM) as a new technology has become a field of research as a result of continuous improvement in numerous approaches in statistics, exploring hidden data in educational environment. An application associated with EDM is a predictive system that can be deployed in early prediction of student academic performance. The importance is to identify poor performers and provide necessary remediation to avoid school drop outs and also encourage high performers. This paper explores certain features of a population of 103 first year students majoring in Computer Science at University of Nigeria, Nsukka. Due to the high number of predicting variables determining student's performance, it is necessary to apply feature selection mechanism using rapid miner to filter these variables. Decision tree, a Machine Learning Algorithm (MLA) was used in training and testing. It was observed that the accuracy is dependent on the datasets on which the model is trained. Two dissimilar datasets achieve different accuracy on the same algorithm. This leads us to conclude that the greatest factor in achieving higher accuracy is the type of datasets not actually the type of classification algorithm.