<![CDATA[This study aims to improve brain tumor MRI image classification performance through selected hyperparameter combinations and ensemble learning. The publicly available dataset consists of 3,264 training images and 394 testing images categorized into four classes: glioma, meningioma, no tumor, and pituitary tumor. The preprocessing stage includes resizing images to 224 × 224 pixels, normalization, and training-data augmentation. Two pre-trained CNN architectures, VGG16 and EfficientNetV2B0, were fine-tuned using selected combinations of the Adam and SGD optimizers, learning rates of 0.0001 and 0.001, and batch sizes of 32. The best configuration was obtained using Adam, a learning rate of 0.0001, and a batch size of 32. VGG16 achieved an accuracy of 90.00%, while EfficientNetV2B0 reached 98.73%. Combining the prediction probabilities of both models using soft-voting ensemble learning increased the accuracy to 99.49%, with two misclassified images. These results indicate that an appropriate training configuration and soft-voting ensemble learning can numerically improve MRI-based brain tumor classification performance.]]>
Copyrights © 2026
