A Comparison of Multi Class Support Vector Machine vs Deep Convolutional Neural Network for Brain Tumor Classification

• Published in: 2023 International Seminar on Application for Technology of Information and Communication (iSemantic) pp. 358 - 363
• Main Authors: Cahyo, Nur Ryan Dwi, Sari, Christy Atika, Rachmawanto, Eko Hari, Jatmoko, Cahaya, Al-Jawry, Rabei Raad Ali, Alkhafaji, Mohamed Ayad
• Format: Conference Proceeding
• Language: English
• Published: IEEE 16-09-2023
• Subjects: Brain Tumor; Classification algorithms; Convolutional neural networks; Deep CNN; Image Classification; Machine Learning; Multi SVM; Support vector machines; Three-dimensional displays; Training; Training data; Transfer learning
• DOI: 10.1109/iSemantic59612.2023.10295336

Multi SVM has long been one of the popular methods in classification, while DCNN has recently gained significant attention in image processing and pattern recognition. This research evaluates the effectiveness of Multi Class Support Vector Machine (M-SVM) and Deep Convolutional Neural Network (DCNN) techniques in classifying brain tumors. A dataset of 2660 3D medical images with dimensions 227 x 227 x 3; including Glioma, Meningioma, and Pituitary tumors, has been partitioned into distinct sets for both training and testing purposes. DCNN approach achieves excellent accuracy in identifying tumor names, with a training accuracy of 97.8% and 100% success rate in 9 experiments. The Multi SVM method demonstrates relatively good accuracy, with training accuracies ranging from 70% to 90% based on different kernel functions. These findings provide valuable insights for selecting appropriate methods in brain tumor classification and encourage further exploration of hybrid Multi SVM-DCNN approaches to enhance accuracy and reliability.