New Multi-View Fuzzy Broad Learning System Enhances Classification

The Broad Learning System (BLS) is a widely used classification method, but it often treats all data points equally, making it vulnerable to noise and outliers in real-world datasets. Additionally, it typically overlooks the geometric structure of data and struggles with multi-source information. To overcome these limitations, a new framework called the Multi-View Graph-Embedded Intuitionistic Fuzzy Broad Learning System (MVGIFBLS) has been introduced. MVGIFBLS integrates several advanced concepts: multi-view learning to combine information from diverse sources, graph embedding to capture intrinsic geometric relationships between samples, and intuitionistic fuzzy theory to enhance robustness against noise. Graph embedding specifically uses local Fisher discriminant analysis to improve class separation. The intuitionistic fuzzy theory, combined with kernel-based neighborhood analysis, further strengthens the model's ability to handle local data structures and noise effectively. Evaluations on various benchmark datasets (UCI, KEEL, AwA) demonstrate that each component of MVGIFBLS contributes positively to its overall performance. The proposed framework consistently achieves higher Area Under the Curve (AUC) scores and maintains robust performance even when exposed to Gaussian feature noise, proving its effectiveness and resilience in complex data classification tasks.