Interpretable Multimodal Classification via Linear Discriminant Tree Ensembles

A new framework uses tree-based ensembles to provide human-understandable explanations for fused text, audio, and visual streams. By clustering concept tokens before routing them through ensembles, the model achieves a balance between high predictive accuracy and the interpretability required for clinical and educational settings.