Scholarlink

Abstract <p>We report an investigation of the influence of potential contamination of a background sample used for training and optimizing a selection by a small fraction of signal events on the multivariate classifier performance. The performance of the boosted decision tree (BDT) classifier trained and optimized on contaminated background sample containing mislabeled signal events is compared to the idealized case of pure training samples. It is demonstrated that the training of the selection is not significantly affected by the absolute number of mislabeled signal events for a given relative background sample contamination value (1000 : 1 background-to-signal ratio). In contrast, the resulting optimization of the selection requirement decreases as the mislabeled signal contribution increases.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Optimization of Rare Event Selection in the Presence of Unextractable Dominant Background

  • 期刊论文

  • 发表日期：

    2025年9月1日

查看全文
  • R. A. Shorkin

摘要

Abstract

We report an investigation of the influence of potential contamination of a background sample used for training and optimizing a selection by a small fraction of signal events on the multivariate classifier performance. The performance of the boosted decision tree (BDT) classifier trained and optimized on contaminated background sample containing mislabeled signal events is compared to the idealized case of pure training samples. It is demonstrated that the training of the selection is not significantly affected by the absolute number of mislabeled signal events for a given relative background sample contamination value (1000 : 1 background-to-signal ratio). In contrast, the resulting optimization of the selection requirement decreases as the mislabeled signal contribution increases.