Two-layer Hybrid XGBoost based on Sample Nearest Neighbor Enveloping Transposition Projection Mechanism
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
XGBoost is a widely recognized and efficient ensemble learning algorithm. However, existing methods commonly train classifiers directly on the original samples without effectively considering the correlation information between them, leading to limited accuracy in cases of low sample separability. To address this, we propose a novel ’two-layer hybrid XG-Boost’ model based on the Sample Nearest Neighbor Enveloping Transposition Projection Mechanism (SNNEPM). This model constructs a two-layer sample space to train and fuse two separate XGBoost models. First, the nearest neighbor samples of each current sample are calculated based on manifold distance. Second, a sample envelope is constructed and reconstructed into a new envelope sample via transposition projection. Finally, two XGBoost models are trained on the original and envelope samples, respectively, and their predictions are fused. Experiments on 10 public datasets show that our method improves ACC by an average of 2% compared to using only the original sample space, and by an average of 1.54% compared to state-of-the-art XGBoost variants, demonstrating its effectiveness.
Fields of Study & Subjects
Machine Learning and Data Classification
Imbalanced Data Classification Techniques
Machine Learning and ELM
Publication details
- Journal / source
- International Conference on Communication Networks and Machine Learning (CNML)
- Pages
- 1012–1016
- Publication date
- 30 January 2026
- Language
- EN