<![CDATA[Purpose – This study aims to compare the performance of MobileNet-V2 and DenseNet121 in classifying brain tumor types from MRI images under identical preprocessing, partitioning, and training conditions. Design/methods/approach – The study used a Kaggle-based dataset consisting of 3,264 MRI images, divided into 88% training data and 12% testing data. Both models were implemented using transfer learning and fine-tuning. Preprocessing included image resizing, normalization, and data augmentation through rotation, flipping, and zooming. The models were trained using the Adam optimizer, a learning rate of 0.0001, batch size of 32, and early stopping. Performance was evaluated using confusion matrix analysis, precision, recall, and F1-score. Findings – The results show that MobileNet-V2 achieved better overall performance than DenseNet121 in brain tumor classification. MobileNet-V2 produced more stable classification results and higher evaluation scores across most tumor classes, particularly in glioma and pituitary tumor prediction. In contrast, DenseNet121 showed a greater tendency to overfit, although both models performed well in identifying non-tumor images. Research implications/limitations – The study is limited by the relatively small dataset size, the use of a single dataset source, and the absence of external validation, which may affect generalizability. Originality/value – This study provides a direct comparative analysis of MobileNet-V2 and DenseNet121 for four-class brain tumor classification and highlights MobileNet-V2 as a more efficient and reliable model for this task.]]>
Copyrights © 2026
