Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Diversity Conscious Refined Random Forest

Created by
  • Haebom

Author

Sijan Bhattarai, Saurav Bhandari, Girija Bhusal, Saroj Shakya, Tapendra Pandey

Outline

In this paper, we propose a novel Random Forest classifier that removes features with little information and selectively generates uncorrelated trees to address the high inference cost and model redundancy issues of Random Forest (RF), which is widely used in various fields for its robust classification performance. The proposed model is iteratively improved by removing features with little information, analytically determining the number of new trees, and removing redundant trees through correlation-based clustering. Experimental results using eight different benchmark datasets (including binary and multi-class) show that the proposed model achieves better accuracy than the standard RF.

Takeaways, Limitations

•
Takeaways:
◦
We show that the inference cost and model complexity of random forests can be reduced by removing features with little information and trees with high correlation.
◦
Achieves improved classification accuracy over standard random forests.
◦
The proposed method demonstrates applicability to various datasets.
•
Limitations:
◦
Additional experiments are needed to determine whether the proposed model's performance improvements are consistent across all datasets.
◦
Further research is needed on the generalizability of the criteria for judging the correlation between features with low information content and trees.
◦
Lack of analysis of the computational complexity of the proposed method.
👍
Made with Slashpage