Komparasi Kernel SVM Pada Klasifikasi Citra Tumor Otak Menggunakan Ekstraksi Fitur GLCM

Abstract

The human brain contains more than 100 billion cells that function to transmit information. Brain tumors occur due to the abnormal growth of cells within or around the brain. Early detection is crucial to prevent tumors from progressing to more severe stages. Manual processes such as biopsies are time-consuming and prone to errors. Therefore, computer technology, such as machine learning, offers an effective solution to support brain tumor diagnosis. This study applies brain tumor image classification using the GLCM feature extraction method and the SVM algorithm as a classifier with four different kernels: linear, polynomial, RBF, and sigmoid with default parameter. The dataset used consists of 2,800 images divided into two classes: tumor and no-tumor. Testing results indicate that the linear kernel achieved an accuracy of 97%, while the polynomial kernel also reached 97%. The RBF kernel achieved 75% accuracy, while the sigmoid kernel produced the lowest accuracy at 48%. These results demonstrate that the linear and polynomial kernels are better at capturing patterns compared to other kernels for brain tumor image classification. With a high confusion matrix score, the linear kernel is recommended for similar image classification tasks.