New deep clustering ensembles tackle imbalanced tabular data

This paper explores the application of unsupervised deep clustering techniques to imbalanced tabular data, a domain where class imbalance typically hinders supervised classification. The researchers introduce two novel ensemble methods for deep clustering: one that aggregates assignments across different embedding dimensions and another that uses majority voting among top-performing algorithms. Experiments on 16 imbalanced tabular datasets demonstrated that these ensemble approaches generally outperform individual deep clustering methods in accuracy, normalized mutual information, and adjusted Rand index scores, offering a robust alternative to supervised classification when class labels are unavailable. AI

IMPACT Offers a potential alternative to supervised classification for imbalanced datasets, improving representation learning without labels.

RANK_REASON The cluster contains an academic paper detailing new methods for unsupervised deep learning on tabular data. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• clustering algorithm
• deep learning
• Hugging Face
• tabular data
• Unsupervised Deep Clustering

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →