SVM method performs a non-linear mapping of original data space into a high-dimensional feature space. The construction of linear discrimination function is useful for replacing the non-linear function in the original data space. This paper aims to efficiently explore the accuracy of SVM with the feature selection method. The selected feature selection method is Correlation-based Feature Selection (CFS), due to the approach's simplicity and speed. This research used an illiteracy rate dataset in Indonesia. The research result showed that the optimised method has overcome the original SVM, with 94 % of accuracy.