Derio001/lri-prediction-chad

Predicting Lower Respiratory Infection (LRI) in children in Chad using machine learning and DHS (Demographic and Health Survey) data. # 🫁 LRI Risk Prediction in Chadian Children Under Five > **Lower Respiratory Infection (LRI) is one of the leading causes of death in children under five in Chad.** > This project builds a machine learning pipeline to predict LRI risk at the individual child level — enabling > targeted health interventions where resources are most scarce. --- ## 🌍 Why This Matters Chad has one of the highest child mortality rates in the world. LRI alone accounts for a significant share of under-five deaths, yet healthcare infrastructure remains severely limited outside N'Djamena. **The core question this project answers:** *"Given a child's demographic profile, household conditions, and nutritional status — how likely are they to develop a Lower Respiratory Infection?"* Answering this with data can help NGOs, ministries of health, and field workers **prioritize interventions** before a child gets critically ill. --- ## 📊 Dataset - **Source**: DHS Program — Chad Standard DHS 2014 - **File**: Children's Recode (KR) — household survey of mothers and children under 5 - **Size**: National representative sample across all regions of Chad - **Key Variables**: Child age/sex, nutritional indicators (WHZ, BMI), mother's education, household wealth, access to healthcare, vaccination status, water/sanitation conditions --- ## 🔬 Methodology The pipeline is designed to be **leak-free and reproducible**: ``` Raw DHS Data ↓ Target Construction (H31, H31B, H31C columns → binary LRI label) ↓ Train/Test Split (Stratified 70/30 — split BEFORE any feature engineering) ↓ Feature Screening (Drop >60% missing | T-test + Chi-Square selection) ↓ Preprocessing (Median/mode imputation | dummy encoding | "Don't know" handling) ↓ SMOTE Balancing (Synthetic oversampling on training set ONLY) ↓ Model Benchmarking (8+ algorithms evaluated) ↓ Evaluation (Accuracy, Recall, Precision — with focus on Recall for health context) ``` --- ## 🤖 Models Benchmarked | Model | Notes | |---|---| | Logi …