Scholarlink

Diabetes can lead to several long-term severe health complications; more than 199 million women are affected by diabetes, with projections indicating a rise to 313 million by 2040. Feature Tokenizer Transformer (FT-Transformer) is proposed to predict diabetes in 1858 women and evaluate its performance with Tabnet, XGboost, RF, and LR. SHAP and LIME methods are used for interpreting FT-Transformer’s performance. FT - Transformer achieved 93.54%, RF (97.31%), XGboost and Tabnet (96.5%), LR (77.42%). FT - Transformer costs more computationally than others; however, it can potentially apply for disease prediction and classification with features containing (text, image, and tabular data) in the medical field. Through SHAP and LIMEthod, an interactive relationship between glucose and pregnancy features in contributing to diabetes prediction was found; glucose is the most predictor of diabetes.

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

Transformer-Based Models for Predicting High-Risk Diabetes in Women Using Tabular Data

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Nguyen Tho,
  • Quoc-Thông Nguyen,
  • Tran Kim Duc,
  • Ludovic Koehl,
  • Tran Kim Phuc

摘要

Diabetes can lead to several long-term severe health complications; more than 199 million women are affected by diabetes, with projections indicating a rise to 313 million by 2040. Feature Tokenizer Transformer (FT-Transformer) is proposed to predict diabetes in 1858 women and evaluate its performance with Tabnet, XGboost, RF, and LR. SHAP and LIME methods are used for interpreting FT-Transformer’s performance. FT - Transformer achieved 93.54%, RF (97.31%), XGboost and Tabnet (96.5%), LR (77.42%). FT - Transformer costs more computationally than others; however, it can potentially apply for disease prediction and classification with features containing (text, image, and tabular data) in the medical field. Through SHAP and LIMEthod, an interactive relationship between glucose and pregnancy features in contributing to diabetes prediction was found; glucose is the most predictor of diabetes.