Release Notes – STB Risk Scoring v1.0

Overview

The STB Risk Scoring package provides an end-to-end scoring pipeline for predicting Suicide Thought and Behavior (STB) risk using a pre-trained ensemble of XGBoost models. The tool standardizes input data, generates risk scores, and assigns percentile-based risk groups to support downstream risk stratification and analysis.

Features

• Automated scoring using pre-trained XGBoost models.
• Built-in feature normalization using the scaler stored in model.pkl.
• Automatic removal of model-excluded variables (dropvars).
• Ensemble prediction with averaged risk scores across multiple models.
• Percentile-based risk group assignment for easy interpretation.
• Simple command-line execution with minimal configuration.

Included Files

• run_score.py – Main scoring script.
• model.pkl – Serialized model package containing trained models and normalization parameters.
• dataset.csv – Input dataset template.
• requirements.txt – Required Python dependencies.

Input Requirements

• Input file must be named dataset.csv.
• Dataset must contain an ID column.
• Feature columns must match those expected by the trained model.

Output

The scoring process generates:

• ID – Record identifier from the input dataset.
• Y_score – Predicted STB risk score.
• RISK_GRP – Percentile-based risk category.

Usage

Install dependencies:

pip install -r requirements.txt

Run scoring:

python run_score.py

Notes

• Risk groups are assigned based on score percentiles, with the top percentile representing the highest predicted risk.
• Users may optionally export results to a CSV file by adding:

df_score.to_csv('scored_output.csv', index=False)

Version 1.0

Initial release of the STB Risk Scoring pipeline.