A Comparative Study of Feature Selection Techniques for Classify Student Performance

Student performance classification is a challenging task for teacher and stakeholder for better academic planning and management. Data mining can be used to find knowledge from student data to improve the performance of classifying model. Before applying a classification model, feature selection method is proposed in data preprocessing process to find out the most significant and intrinsic features. In this research, we propose a comparison of four feature selection methods: genetic algorithms, support vector machine, information gain, and minimum redundancy and maximum relevance with four supervised classifiers: naive bays, decision tree, k-nearest neighbor, and neural network. The experimental results show that the minimum redundancy and maximum relevance feature selection method with 10 feature selected give the best result on 91.12% accuracy with a k-nearest neighbor classifier. The result of the present study shows that the advantage of future selection to find a minimum and significant of feature is more effective to classify the student performance.