Multi-class brain tumor diagnosis using MRI: A dynamic reinforcement ensemble learning approach

Early and accurate diagnosis is essential for effective clinical decision-making, particularly when working with complex and imbalanced medical datasets. Traditional static ensemble models often struggle to adapt to such challenges, limiting their generalization and performance.

This study aims to develop a novel dynamic ensemble learning framework that enhances diagnostic accuracy by addressing the limitations of static ensemble strategies through adaptive model selection and dynamic weight adjustment.

We propose a Dynamic Reinforcement Ensemble Learning Model that leverages reinforcement learning (RL) to dynamically select the most suitable base classifiers and adjust their contribution weights based on input data characteristics. The model was evaluated on three benchmark medical datasets: LSBTDK-DAT, FIGSHARE-DAT, and THYROID-DAT. Comparative analyses and ablation studies were conducted to assess performance gains and the impact of each dynamic component.

On LSBTDK-DAT, the proposed model achieved an accuracy of 99.55% and an F1-score of 99.54%. On THYROID-DAT, it reached 99.35% accuracy and an F1-score of 98.89%. Across all datasets, the model outperformed existing state-of-the-art methods by up to 7% in accuracy and 5% in F1-score. Ablation experiments confirmed that the combination of dynamic selection and dynamic weighting consistently produced the best outcomes.

The integration of dynamic reinforcement ensemble selection with adaptive weighting enables the model to robustly handle diverse and complex medical data. These results demonstrate the model's potential for intelligent clinical decision support systems and lay the foundation for scalable, high-precision medical AI solutions.