New LCEN algorithm and diffMCC loss boost classification task performance

Researchers have developed a modified LASSO-Clip-EN (LCEN) algorithm specifically for classification tasks, maintaining its interpretability and feature selection capabilities. Experiments show this new LCEN consistently achieves high macro F1 scores and Matthews correlation coefficients (MCC), outperforming most other models tested and eliminating an average of 56% of input features. Additionally, a novel weighted focal differentiable MCC (diffMCC) loss function was evaluated, demonstrating that models trained with it consistently outperformed those trained with weighted cross-entropy loss, achieving significantly higher F1 scores and MCCs. AI

IMPACT Introduces novel feature selection and loss function techniques that could improve classification model performance and interpretability.

RANK_REASON Academic paper introducing a new algorithm and loss function for classification tasks.

Read on arXiv cs.LG →

• LASSO-Clip-EN
• Matthews correlation coefficient
• weighted cross-entropy loss
• weighted focal differentiable MCC loss function
• LCEN

• paper
• other

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