Predicting Performance of Primary School Students and Determining The Dominant Factors Using Data Mining with Feature Selection Methods

Abstract

Educational organizations consistently attempt to investigate students' behaviors in learning and give early predictions for intervening and improving their learning performance, especially in primary education, which is the core stone of education. Educational data mining offers different practical prediction algorithms to predict student performance and various feature selection methods to determine the dominant factors which affect student performance. This study applied three data mining algorithms, K-Nearest Neighbor, Naive Bayes, and Decision Tree, to predict the performance of students based on their socio-demographic characteristics. The feature selection methods Information Gain Attribute Evaluation, Wrapper Subset Evaluation, and Classifier Subset Evaluation were used to identify the dominant factors. The research was done by adopting the CRISP methodology using a data set obtained from one of Bahrain's public primary boys' schools. The results showed that the model of the k-Nearest Neighbors algorithm with Wrapper Subset Evaluation as a feature selection method had achieved the highest accuracy rate of 67.6%. For the dominant factors, Student Age and Student Level have the most impact on student performance, followed by Parent education level.