Survival Analysis With CatBoost

This repository contains a comprehensive survival analysis project focusing on mortality prediction and waitlist analysis in healthcare. The project utilizes various machine learning techniques, including CatBoost, XGBoost, and Random Forest models, to analyze and predict patient outcomes.

Project Structure

graph LR
  A[survival_analysis] --> B[final_model]
  B --> B1[survival_analysis_final_catboost_model.qmd]
  B --> B2[survival_analysis_final_catboost_model.html]
  B --> B3[stratified_risk_metrics.csv]
  B --> B4[sorted_shap_values_interactive.html]
  B --> C[images]
  C --> C1[catboost_beeswarm.png]
  C --> C2[mean_absolute_shap_importance.png]
  C --> C3[partial_dependence_plots_test]
  C --> C4[partial_dependence_plots_train]
  C --> C5[shap_correlation_plot.png]
  C --> C6[shapley_bar_one_hot.png]
  A --> D[tripod_ai_checklist]
  D --> D1[tripod_ai_checklistt.qmd]
  D --> D2[tripod_ai_checklist.html]
  A --> F[model_data]
  F --> F1[center-stats.qmd]
  F --> F2[center-stats.html]
  F --> F3[model_data.qmd]
  F --> F4[model_data.html]
  F --> F5[waitlist-data2.qmd]
  F --> F6[waitlist-data2.html]
  A --> G[eda]
  G --> G1[data_profile_report.html]
  G --> G2[listing-mortality-prediction.qmd]
  G --> G3[listing-mortality-prediction.html]

Loading

Project Components

Final Model

The final_model/ directory contains:

• The final CatBoost model for survival analysis: survival_analysis_final_catboost_model.qmd
• Interactive Shap Values Feature Importance visualization Shap Values
• Stratified Accuracy Risk Metrics
• images directory

Images

The images/ directory stores generated plots and figures:

• CatBoost beeswarm plot
• Mean absolute SHAP importance
• Partial dependence plots - Test and Train
• SHAP correlation plot
• Shapley bar plot (one-hot encoded)

TRIPOD+AI Checklist

The tripod_ai_checklist/ directory contains the completed TRIPOD+AI Checklist.

Model Data

The model_data/ directory contains the data pre-processing and processing used in the analysis.

• Listing Center specific analysis
• Candidate pre-processing and candidate wait list specific analysis (censoring)
• Final Model Data Set (structure_plot.json)

Exploratory Data Analysis (EDA)

The eda/ directory contains a data profiling report and initial model evaluations:

• DataExplorer dataset profile report data_profile_report.html
• Comparison of Logistic Regression, GAM, Random Forest, and XGBoost models listing-mortality-prediction.html

Usage

To reproduce the analysis:

1. Ensure you have R and the required packages installed.
2. Request the data files and place in /data directory.
3. Run survival_analysis_final_catboost_model.qmd.

Contact

Jerome Dixon