How codingmermaid/hospital-readmission-ml works
This repository is an educational 30-day hospital-readmission prediction project: it builds and evaluates models, then serves a risk estimate through a Streamlit app. The main user flow is patient and encounter details → engineered feature row → LightGBM probability and threshold-based risk flag. The README documents the broader cleaning, EDA, baseline-model, evaluation, and SHAP workflow. Some pipeline links reflect documented stages; the data files are not included in the tree, and source excerpts do not establish script orchestration.
Main entry points: lightgbm_model.py LightGBM training · app.py Streamlit risk estimator · clean.py Data cleaning · feature_engineering.py Feature engineering
Components 14
Risk estimation app
- Streamlit risk estimator
app.py - LightGBM model
model/lightgbm_model.joblib - Categorical features
model/categorical_cols.joblib - Decision threshold
model/lightgbm_threshold.joblib
Data preparation
- Data cleaning
scripts/clean.py - Exploratory charts
scripts/visualize.py - Feature engineering
scripts/feature_engineering.py
Model development
- Logistic baseline
scripts/baseline_model.py - LightGBM training
scripts/lightgbm_model.py
Evaluation and interpretation
- ROC and PR evaluation
scripts/roc_pr_curves.py - Accuracy comparison
scripts/accuracy_trap_chart.py - SHAP analysis
scripts/shap_analysis.py
Other components
- App user
- UCI encounter data
Connections 19
- App user → Streamlit risk estimator: enters details
README.md - Streamlit risk estimator → LightGBM model: loads model
app.py - Streamlit risk estimator → Categorical features: loads feature list
app.py - Streamlit risk estimator → Decision threshold: loads cutoff
app.py - Streamlit risk estimator → LightGBM model: scores row
app.py - Streamlit risk estimator → Decision threshold: compares score
app.py - UCI encounter data → Data cleaning: supplies encounters
scripts/clean.py - Data cleaning → Exploratory charts: explores cleaned data
scripts/visualize.py - Data cleaning → Feature engineering: provides cleaned data
scripts/feature_engineering.py - Feature engineering → Logistic baseline: provides model features
scripts/baseline_model.py - Feature engineering → LightGBM training: provides model features
scripts/lightgbm_model.py - LightGBM training → LightGBM model: saves model
scripts/lightgbm_model.py - LightGBM training → Categorical features: saves feature list
scripts/lightgbm_model.py - LightGBM training → Decision threshold: saves cutoff
scripts/lightgbm_model.py - Logistic baseline → ROC and PR evaluation: compared in curves
scripts/roc_pr_curves.py - LightGBM training → ROC and PR evaluation: compared in curves
scripts/roc_pr_curves.py - Logistic baseline → Accuracy comparison: included in comparison
README.md - LightGBM training → Accuracy comparison: included in comparison
README.md - LightGBM training → SHAP analysis: explained by
scripts/shap_analysis.py