使用共识定义进行间接数据标注的临床机器学习中的有效性问题
Validity problems in clinical machine learning by indirect data labeling using consensus definitions
AI总结:
该研究指出临床机器学习中因训练标签由间接测量确定且基础测量值纳入输入导致的有效性问题,提出识别问题数据集与黑盒模型的通用流程,并在脓毒症早期预测任务中验证。
摘要:
我们揭示了机器学习在医学疾病诊断这一重要应用领域中存在的一个有效性问题。当训练数据中的目标标签由间接测量确定,且确定该间接测量所需的基础测量值被纳入输入数据表示时,就会出现这一问题。基于此类数据训练的机器学习模型只会学习精确复现已知的目标定义,除此之外学不到任何其他内容。这类模型在结构相似的测试数据上表现完美,但在定义性基础测量值不可用或仅部分可用的真实场景中会彻底失效。我们提出了一种通用流程,可识别存在问题的数据集以及基于这些数据集训练的黑盒机器学习模型,并在脓毒症早期预测任务上示例验证了我们的检测流程。
英文摘要:
We demonstrate a validity problem of machine learning in the vital application area of disease diagnosis in medicine. It arises when target labels in training data are determined by an indirect measurement, and the fundamental measurements needed to determine this indirect measurement are included in the input data representation. Machine learning models trained on this data will learn nothing else but to exactly reconstruct the known target definition. Such models show perfect performance on similarly constructed test data but will fail catastrophically on real-world examples where the defining fundamental measurements are not or only incompletely available. We present a general procedure allowing identification of problematic datasets and black-box machine learning models trained on them, and exemplify our detection procedure on the task of early prediction of sepsis.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。