Scholarlink

The world will spend more than $10 trillion on healthcare by 2026 and therefore there is a strong need for credible and transparent frameworks to estimate costs that enable responsible use of resources. Towards addressing this, the eHealth-XAI scoring mechanism uses an AI explainable approach for predicting individual medical costs using the Kaggle Medical Cost Personal Dataset. Decision Tree, and Random Forest achieving an R-squared value of 0.85 (95% CI: [0.83–0.87]) and a Mean Absolute Error of $1200 (95% CI: [1100–1300]). Two interpretable methodologies (SHAP at general level of factor attribution and LIME for specific explanation by instance) are applied to guarantee transparency. An exploratory user-facing tool is presented via a Streamlit environment to serve the community stakeholders. Given the limited number of observations in this trial (1,338), our proposed framework serves as a proof-of-concept for validation on the larger scale in real hospital system with an aim to facilitate credible decision-making in digital healthcare.

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

Interpretable Healthcare Cost Prediction Using Explainable XGBoost with SHAP and LIME

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • G. R. Ramya,
  • M. Vamsi Krishna,
  • Sai Tanvika Nori,
  • K. P. Yeshwanth,
  • G. K. N. V. Jagannadh

摘要

The world will spend more than $10 trillion on healthcare by 2026 and therefore there is a strong need for credible and transparent frameworks to estimate costs that enable responsible use of resources. Towards addressing this, the eHealth-XAI scoring mechanism uses an AI explainable approach for predicting individual medical costs using the Kaggle Medical Cost Personal Dataset. Decision Tree, and Random Forest achieving an R-squared value of 0.85 (95% CI: [0.83–0.87]) and a Mean Absolute Error of $1200 (95% CI: [1100–1300]). Two interpretable methodologies (SHAP at general level of factor attribution and LIME for specific explanation by instance) are applied to guarantee transparency. An exploratory user-facing tool is presented via a Streamlit environment to serve the community stakeholders. Given the limited number of observations in this trial (1,338), our proposed framework serves as a proof-of-concept for validation on the larger scale in real hospital system with an aim to facilitate credible decision-making in digital healthcare.