ClarusC64/clinical-renal-filtration-instability-v0.1

该数据集用于评估模型是否能检测肾脏过滤和废物清除的不稳定性。每一行代表一个简化的肾功能监测场景，包含三个时间点的数据。任务是根据提供的数据（包括肌酐轨迹、估计过滤代理、尿量轨迹、血尿素氮代理、代谢废物代理和干预延迟等信号）判断肾脏过滤是保持稳定（label=0）还是趋向不稳定（label=1）。数据集还包含干扰变量（实验室噪声和图表噪声）。评估方法包括准确率、精确率、召回率、F1分数、混淆矩阵和数据集完整性诊断。数据集是Clarus稳定性推理基准的一部分，采用MIT许可证。

This dataset evaluates whether models can detect instability in renal filtration and waste clearance. Each row represents a simplified kidney function monitoring scenario observed across three time points. The task is to determine whether renal filtration remains stable (label=0) or is moving toward renal instability (label=1) based on provided data, including signals such as creatinine trajectory, estimated filtration proxy, urine output trajectory, blood urea nitrogen proxy, metabolic waste proxy, and intervention delay. The dataset also includes decoy variables (lab_noise and chart_noise). Evaluation metrics include accuracy, precision, recall, f1, confusion matrix, and dataset integrity diagnostics. The dataset is part of the Clarus Stability Reasoning Benchmark and is licensed under MIT.