Weighted Majority Voting Ensemble for MRI-Based Brain Tumor Classification Using Capsule Networks and XGBoost

Brain Tumor (BT) is the most serious illness affecting humans, and its diagnosis is a complex process. Tumor location and type significantly affect treatment decisions, and survival rates improve with accurate identification and classification in the early stages. Magnetic Resonance Imaging (MRI) is mainly used for brain tumor analysis, but manual detection and classification by clinicians is challenging, often leading to high error rates, inaccurate diagnoses, and prolonged time requirements. To overcome these challenges, this paper introduces a novel hybrid classification approach that combines Capsule Networks (CapsNet) and XGBoost (XGB) to classify brain tumors from MRI images. The preprocessing step includes normalization, image blurring, resizing, contrast enhancement, and noise elimination, which are used to improve image quality. The classification process employs CapsNet to capture hierarchical features and spatial relationships in the images, while XGB utilizes extracted features, such as texture, intensity, and shape, to classify tumors effectively. To improve diagnostic accuracy, a Meta Ensemble Model combines the predictions of both algorithms using a Weighted Majority Voting approach, adjusting contributions based on each model’s confidence. Additionally, the Mantis Search Algorithm (MSA) is utilized for hyperparameter tuning, optimizing model performance by exploring the hyperparameter space effectively. The experiment assessed using the Brain Tumor MRI Dataset and Figshare Brain Tumor Dataset demonstrates the effectiveness of the proposed method, achieving an accuracy of 99.34% and a precision of 98.82%. These results indicate that the hybrid method is highly effective in accurately classifying various brain tumor types, which provides the best solution for clinical diagnostics.