LFS-FRAME: Leakage-Free Stacked Ensemble for Multiclass Classification
A new machine learning framework, LFS-FRAME, combines Kolmogorov-Arnold Networks (KAN) and XGBoost for robust multiclass classification. The method addresses challenges like high inter-class similarity, class imbalance, and data distribution variability. It uses a strict out-of-fold stacking strategy to prevent data leakage and construct unbiased meta-features. The approach integrates functional learning from KAN and rule-based learning from XGBoost, aiming to overcome limitations of each method alone. The paper is available on arXiv with ID 2607.22081.
Key facts
- LFS-FRAME is a leakage-free stacked ensemble framework.
- It integrates KAN for functional learning and XGBoost for rule-based learning.
- Uses strict out-of-fold stacking to prevent data leakage.
- Addresses multiclass classification challenges: high inter-class similarity, class imbalance, data distribution variability.
- KAN captures smooth functional relationships; XGBoost handles structured features.
- Neural networks can overfit; rule-based classifiers limit smooth function capture.
- Paper available on arXiv: 2607.22081.
- Published as arXiv:2607.22081v1.
Entities
Institutions
- arXiv