getachewgetu/stunting-risk-model

Stunting Risk Scorer — Rwanda

Logistic Regression model for predicting childhood stunting risk at household level. Trained on synthetic NISR-style data for the AIMS KTT Hackathon (S2.T1.2).

Model Performance

• Test AUC-ROC: 0.9689
• Accuracy / Precision / Recall / F1: 0.8667
• Best hyperparameters: C=10.0 (L2 regularisation)

Features

Feature	Description
avg_meal_count	Average meals per day
water_source_enc	Water source quality (0=best, 4=worst)
sanitation_tier_enc	Sanitation level (0=best, 3=worst)
income_band_enc	Income band (0=high, 2=low)
children_under5	Number of children under 5
meal_x_water	Interaction: meal_count × water_source_enc
deprivation_index	Sum of water + sanitation + income encodings

Usage

from huggingface_hub import hf_hub_download
import joblib

path = hf_hub_download(repo_id="getachewgetu/stunting-risk-model", filename="risk_model.pkl")
artifact = joblib.load(path)
model = artifact['model']
scaler = artifact['preprocessor']

Training

Trained with 5-fold stratified CV, GridSearchCV hyperparameter tuning. See train_model.py for the full pipeline.