Feature Extraction and Dimensionality Reduction in Multiclass Brain Tumor Classification

Classifying brain tumors in medical imaging is a complex task due to the visual similarities among different tumor types. This study explores the use of Support Vector Machine (SVM) as the primary classifier, combined with feature extraction method and a dimensionality reduction technique, to enhance classification accuracy. Specifically, Oriented FAST and Rotated BRIEF (ORB) is employed for feature extraction, while Principal Component Analysis (PCA) reduces the feature space. The dataset comprises three categories: glioma, meningioma, and no tumor. On a test set of unseen MRI images, the proposed method achieved 95.88% accuracy(acc_val), with corresponding F1-score(f1_val) of 94.5%, precision(prec_val) of 94.2%, and recall(rec_val) of 94.35%. The results demonstrate that SVM, when paired with ORB and PCA, offers a robust and efficient approach for multiclass brain tumor classification.