Enhancing the Classification Performance of Students Behavior on Serious Game using Discretization-based k-NN

Data relating to the serious game interaction in the education area can be mined to find the students’ behavior indicating the understanding of the specific subject. To the best of our knowledge, this is the first research to improve the classification performance of students’ behavior on a serious game using k-NN based on the discretization method. The discretization method applied is unsupervised discretization called equal frequency. Then, we combine on k-NN with Manhattan as a distance metric. Additionally, we evaluate the performance using the cross-validation. Then, we analyze the result using the general classification metric, the sieve diagram, and ROC. The experimental result shows that the combination of k-NN and the discretization method with 5-intervals can achieve the highest level of all metrics and the widest Area Under Curve (AUC). This indicates that this proposed method can improve a higher performance than the k-NN without discretization.