YerevaNN/mimic3-benchmarks

Provides standardized benchmark datasets for four clinically-relevant prediction tasks—mortality, decompensation, length-of-stay, and phenotyping—each mapped to core ML problem types (classification, time series classification, regression, multilabel sequence classification). The toolkit handles end-to-end data transformation from raw MIMIC-III CSVs through subject extraction, event validation, and episode segmentation with integrated outlier detection and variable normalization. Includes baseline RNN models, evaluation metrics, and reader utilities designed for multitask learning architectures on clinical time series data.

874 stars. No commits in the last 6 months.