Exploring the Kernel on SVM to Enhance the Classification Performance of Students' Academic Performance

Information relating to the student's performance is important to teachers to prevent failure in learning achieving. On the other side, online learning produces massive student data. The methods in data mining can be applied to student data to generate this information. However, the previous research does not yet do to explore the kernels in SVM to obtain the best performance of the model. The paper focuses to explore the kernels in SVM to find the suitable kernel in our student data. The experiment using many scenarios is done after the model is built. The experimental result shows that the linear kernel achieves the highest level in both evaluation techniques. On the cross-validation technique and the percentage split, the highest accuracy levels are reached on fold=5 about 88.8%, and on the training set size =60% about 88.28, respectively. in addition, the performance of this linear kernel, in terms of level accuracy. higher by 42.4%, 36.2& and 44.3% compared to the sigmoid, RBF and polynomial kernels on the cross validation technique, respectively. In the percentage split technique, this linear kernel is higher than the sigmoid, RBF and polynomial kernels by 41.5%, 37.7% and 42.4%, respectively. This indicates that our student data more appropriate is approximated by the linear model.