Scholarlink

<p>Tabular data is a common format for storing information in rows and columns, representing data entries and their associated features. Existing deep learning models designed for tabular data often rely on uni-dimensional feature selection mechanisms, which fail to adequately capture contextual dependencies of feature importance. This can result in missed crucial interactions, bias toward high-impact features, and limited generalization in attention generalization. To overcome these challenges, this study introduces a novel iterative feature exclusion module that enhances the estimation of feature importance. The proposed module iteratively excludes a single feature from the input data and computes attention scores that represent the impact of each feature on the prediction. By aggregating the attention scores across iterations, the module generates a refined representation of feature importance that captures both the global and local feature interactions. The effectiveness of the proposed module was evaluated on four public datasets, and the results consistently demonstrated its superior performance over state-of-the-art methods and baseline models in both feature ranking and classification tasks. The code is publicly available at <a href="https://github.com/abaraka2020/Iterative-Feature-Exclusion-Ranking-Module">https://github.com/abaraka2020/Iterative-Feature-Exclusion-Ranking-Module</a> and <a href="https://github.com/mohalim/IFENet">https://github.com/mohalim/IFENet</a>.</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

Iterative feature exclusion ranking for deep tabular learning

  • 期刊论文

  • 发表日期：

    2025年12月16日

查看全文
  • Fathi Said Emhemed Shaninah,
  • AbdulRahman M. A. Baraka,
  • Mohd Halim Mohd Noor

摘要

Tabular data is a common format for storing information in rows and columns, representing data entries and their associated features. Existing deep learning models designed for tabular data often rely on uni-dimensional feature selection mechanisms, which fail to adequately capture contextual dependencies of feature importance. This can result in missed crucial interactions, bias toward high-impact features, and limited generalization in attention generalization. To overcome these challenges, this study introduces a novel iterative feature exclusion module that enhances the estimation of feature importance. The proposed module iteratively excludes a single feature from the input data and computes attention scores that represent the impact of each feature on the prediction. By aggregating the attention scores across iterations, the module generates a refined representation of feature importance that captures both the global and local feature interactions. The effectiveness of the proposed module was evaluated on four public datasets, and the results consistently demonstrated its superior performance over state-of-the-art methods and baseline models in both feature ranking and classification tasks. The code is publicly available at https://github.com/abaraka2020/Iterative-Feature-Exclusion-Ranking-Module and https://github.com/mohalim/IFENet.