Mhya1/ml_imputation_nutrition

Compared ML imputation methods (KNN, MICE, Random Forest) for missing WHZ data in SMART surveys using a DHS-consistent dataset of 1,500 children from Northeast Nigeria. MICE and RF outperformed mean substitution by 25–35% RMSE. Includes reproducible Python notebook and research manuscript. ```markdown # README: ML-Based Imputation of Missing Nutrition Data in Humanitarian Surveys This notebook replicates the analysis presented in the paper: >*'Machine Learning-Based Imputation of Missing Household Nutrition Data in Humanitarian Surveys: A Comparative Study Using DHS-Consistent Data from Northeast Nigeria'* by Joshua Samuel Dauda. ## Overview This study compares various machine learning-based imputation methods for handling missing 'Weight-for-Height Z-score' (WHZ) data, a proxy for acute malnutrition, in humanitarian surveys. The dataset used is a simulated DHS-consistent dataset reflecting the demographic and nutritional structure of Northeast Nigeria. ## Key Sections: 1. **Environment Setup**: Imports necessary libraries and sets up the environment. 2. **Data Generation**: Simulates a DHS-consistent child nutrition dataset. 3. **Introduce Missing Data**: Artificially introduces missing data into the 'whz' variable using three mechanisms: Missing Completely At Random (MCAR), Missing At Random (MAR), and Missing Not At Random (MNAR). 4. **Visualize Missing Patterns**: Illustrates the introduced missing data patterns, particularly by wealth quintile. 5. **Evaluation Function**: Defines functions for setting up different imputers (Mean, KNN, MICE, Random Forest) and evaluating their performance based on RMSE, MAE, R², and Bias. 6. **Run Comparison**: Executes the imputation methods across all three missing data mechanisms and presents the results. 7. **Figure 2 — RMSE Comparison Bar Chart**: Visualizes the RMSE performance of each imputation method for different missingness mechanisms. 8. **Figure 3 — Scatter: True vs Imputed WHZ (MAR)**: Compares true WHZ values with imputed values for Mean and Random Forest imputation under the MAR mechanism. 9. **Figure 4 — GAM Prevalence Estimation Error**: Assesses how accurately each method estimates the prevalence of Global Acute Malnutrition (GAM). 10. **Sensitivity Analysis — …