Classifying Student Achievement Using K-NN Based on Feature Normalization Techniques

The existence of students' achievement classification with high performance is very important to support teachers in the evaluation of the process of learning. However, not much research has addressed this issue. Accordingly, our research concentrate on the establishment of a classification model for students' achievement with high performance. To achieve this goal, we explore three feature normalization techniques (feature standardized, feature scaled, and feature centered), which are combined with the K-NN method. The K-NN parameters are set with Euclidean distances and the number of neighbors 2, 4, and 6.Furthermore, the evaluation technique uses cross-validation with 2, 3, 5, 10, and 20 folds. The experimental results reveal that the combination of both normalization techniques and K-NN produces the best performance in terms of average accuracy, namely: Standard_K-NN and Scale_K-NN about 84.52% by the number of neighbor = 2. This shows that both techniques are more appropriate to the characteristic of our student data, which is indicated by their accuracy average outperforming the Center_K-NN method by 1.2%.