Robust Multiclass Brain Tumor Classification: Leveraging Swin Transformer and Feature Optimization with Ensemble Learning

Brain tumor is one of the most serious health problems, and early discovery is crucial to improving patient outcomes. Magnetic resonance imaging (MRI) is vital in early detection, diagnosis, and timely treatment. Due to the complexity and diversity of multiclass brain tumors, feature extraction becomes more difficult because it requires distinguishing between tumor categories, which typically overlap features in MRI. Traditional diagnostic approaches, while effective, frequently have limits in terms of accuracy, speed, and usability. This work presents a voting-based ensemble framework for multiclass brain tumor classification, which incorporates three variations of a pre-trained Swin Transformer for feature extraction. Among these features, we used only refined features to train machine learning models by applying several feature removal processes. Numerous experiments were conducted to evaluate the performance of the suggested model and compare the outcomes to existing transformer-based models, such as the Vision Transformer and the Swin Transformer. The experimental findings revealed that the model outperformed the other models, with a maximum accuracy of 97.56%.