OPTIMIZATION OF NODE SIZE CONFIGURATION IN CNN-ELM MODEL FOR BRAIN TUMOR MRI IMAGE CLASSIFICATION

Ahmad, Sulthan
Rahmat, Basuki
Tri Anggraeny, Fetty

Jusikom: Jurnal Sistem Informasi Ilmu Komputer
Vol. 8 No. 1 (2024): JUSIKOM: JURNAL SISTEM INFROMASI ILMU KOMPUTER

This study proposed a method to classify four types of brain tumors—Glioma, Meningioma, Pituitary, and Non-Tumor—using the Kaggle Brain Tumor MRI Dataset. The research involved stages of data collection, preprocessing, model design, model training, and evaluation. A hybrid Convolutional Neural Network - Extreme Learning Machine (CNN-ELM) algorithm was employed, demonstrating the importance of selecting the optimal number of hidden nodes for achieving high accuracy. The test results revealed that with 2000 hidden nodes, the CNN-ELM model achieved an overall accuracy of 98.86%, with F1-scores of 97% for Glioma, 98% for Meningioma, 100% for Non-Tumor, and 100% for Pituitary tumors. In comparison, the model with 1000 hidden nodes achieved an accuracy of 96.96%, while models with 3000 and 4000 hidden nodes achieved 98.10% and 96.58% accuracy, respectively. These findings highlight the critical role of hidden node selection in optimizing model performance. The CNN-ELM algorithm proves to be a viable alternative for classifying brain tumor MRI images, contributing to advancements in medical technology.