Scholarlink

In scenarios of intelligent recommendation for diagnostic and treatment plans based on clinical data, the high dimensionality of data significantly limits the system’s rapid response capability and the speed of generating precise recommendations, directly undermining the utility and effectiveness of decision support systems in emergency medical situations. This paper proposes a feature selection framework based on large language models (LLMs), called Initial-LLM. It analyzes feature data and relevant background information to infer feature importance, using the learned knowledge to guide the evolution of meta-heuristic algorithms. Comparative experiments on six medical datasets showed that integrating GPT-4 improved performance, particularly when combined with GA and DE algorithms, achieve performance similar to or even better than TLBO. Although the model’s generalizability on ultra-high dimensional data is unvalidated, Initial-LLM enhances both analysis depth and response speed, offering rapid, accurate clinical decision support.

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

Initial-LLM: A Large Language Model-Guided Metaheuristic Framework for Enhanced Feature Selection in Clinical Decision Support Systems

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Zihang Wang,
  • Ye Liang,
  • Wenwei Sun,
  • Chao Xu,
  • Yining Zhou,
  • Yong Zhang

摘要

In scenarios of intelligent recommendation for diagnostic and treatment plans based on clinical data, the high dimensionality of data significantly limits the system’s rapid response capability and the speed of generating precise recommendations, directly undermining the utility and effectiveness of decision support systems in emergency medical situations. This paper proposes a feature selection framework based on large language models (LLMs), called Initial-LLM. It analyzes feature data and relevant background information to infer feature importance, using the learned knowledge to guide the evolution of meta-heuristic algorithms. Comparative experiments on six medical datasets showed that integrating GPT-4 improved performance, particularly when combined with GA and DE algorithms, achieve performance similar to or even better than TLBO. Although the model’s generalizability on ultra-high dimensional data is unvalidated, Initial-LLM enhances both analysis depth and response speed, offering rapid, accurate clinical decision support.