Fairness-aware ICU mortality prediction using MIMIC-III data and Group-Aware SMOTE

A Multimodal Deep Learning framework for early ICU outcome prediction (Length of Stay and Mortality) on MIMIC-III. Features bi-LSTM vital-sign encoding, an MLP for clinical + lab data, 8-head Cross-Modal Attention Fusion, Platt temperature scaling calibration, and an interactive Streamlit clinical decision support dashboard.

Predicts ICU patient survival using clinical and demographic features with logistic regression, decision trees, LASSO, and AIC-based model selection in R. Designed for healthcare decision-making with real-world medical insights drawn from 92K+ patient records.

Explainable deep learning framework for ICU mortality prediction using latest MIMIC-IV v3.1 with SHAP/LIME based clinical decision support.

Reproducible pipeline for SUPPORT2 survival analysis with three-layer leakage audit, LODGO transportability benchmark, and bootstrap-validated baselines. PLOS ONE 2026.