clinical-ml

🧪Yet Another ICU Benchmark: a holistic framework for the standardization of clinical prediction model experiments. Provide custom datasets, cohorts, prediction tasks, endpoints, preprocessing, and models. Paper: https://arxiv.org/abs/2306.05109

Repository for the Paper: „On the Importance of Step-wise Embeddings for Heterogeneous Clinical Time-Series“

SA-AKI Mortality Prediction — Survival analysis & binary classification for Sepsis-Associated Acute Kidney Injury using MIMIC-IV. CatBoost, LightGBM, XGBoost, Logistic Regression. AUROC ~0.80+

Early sepsis risk prediction pipeline using machine learning on ICU clinical data.

Stroke risk prediction using Random Forest in R — ROSE class balancing, 14.03% OOB error rate, confusion matrix evaluation & deployed as .rds model for clinical use

Foundation model for health trajectory prediction — architecture shootout across Transformer, Mamba, and Continuous-Time

Automated seizure onset detection on intracranial EEG (SWEC-ETHZ): feature extraction, EDA, and classical ML baselines with leave-one-patient-out evaluation

Hospital readmission prediction ML pipeline in R — 6 algorithms (LASSO, Random Forest, SVM, KNN, Naive Bayes, Decision Tree) on 69,984 diabetic patients using Tidymodels + SMOTE

Predicts 30-day hospital readmission risk in diabetic patients using ICD-9 code grouping, random oversampling, and HistGradientBoosting on 101,766 clinical encounters. Best model: F1 = 0.2818, AUC = 0.6704.

x402-protected REST API for MIMIC-IV clinical dataset schemas, stats, and previews

Early longer-stay risk for the renal cohort — a leakage-audited, calibrated, externally-validated clinical-ML case study (dbt + DuckDB + scikit-learn, MIMIC-IV + eICU).

Machine learning pipeline for multi-class treatment prediction in lung adenocarcinoma (LUAD) using patient-level molecular profiles, featuring ensemble-based model aggregation, benchmarking across diverse classifier architectures, and systematic performance evaluation.

Multi-module clinical ML framework for real-time ICU mortality prediction, multimorbidity risk scoring, and waveform event detection with automated MLOps drift monitoring and ensemble fusion.

Early sepsis prediction in ICU patients using bidirectional LSTM — flags at-risk patients 6-12 hours before clinical diagnosis

End-to-end ML pipeline for predicting 30-day hospital readmission in diabetic patients — 8 models, SHAP explainability, clinical interpretation | UCI 130-US Hospitals dataset

Leakage-aware MATLAB framework for early sepsis risk research using MIMIC-III Demo and patient-level validation

Classificazione spiegabile di radiografie del torace con incertezza (MC-Dropout), calibrazione e Grad-CAM; uv + Streamlit

Production ML pipeline for clinical decision support - risk scoring, treatment optimization, outcome prediction

Cardiac risk prediction Streamlit app — XGBoost/RF on Cleveland dataset biomarkers, calibrated probability output, SHAP attribution, ROC curves, AUC-optimised threshold.

End-to-end ML demo for clinical risk scoring on synthetic tabular patient data (data → features → models → metrics → FastAPI /predict).