ChiaHsuan-Wu/Predicting-IPC-Phase-3-Food-Insecurity

An end-to-end data science pipeline using machine learning (XGBoost, Random Forest) to predict regional IPC Phase 3+ food insecurity across 5 African nations. # Predicting IPC Phase 3+ Food Insecurity **MSc Data Science Project | In Collaboration with Zero Hunger Lab & Tilburg University** ### Project Overview This repository contains the end-to-end predictive analytics pipeline developed to forecast **IPC Phase 3+ (Crisis or worse) food insecurity conditions** across 5 sub-Saharan African nations (Central African Republic, Madagascar, Mozambique, Somalia, and South Sudan) using open-source, public indicators. Leveraging multi-domain spatial-temporal datasets provided in collaboration with Tilburg University's **Zero Hunger Lab**, this project successfully applies cost-sensitive machine learning to transform volatile environmental and conflict data into high-recall, policy-relevant humanitarian insights, contributing directly to the UN Sustainable Development Goal 2 (SDG2). ### Tech Stack & Methodology - **Language & Environment:** Python, Jupyter Notebook - **Libraries:** Pandas, NumPy, Scikit-Learn, XGBoost, Matplotlib, Seaborn, SHAP - **Validation Framework:** Chronological TimeSeriesSplit Cross-Validation (K=5) with strict time-based Train/Test split to prevent temporal data leakage. - **Imbalance Mitigation:** Cost-sensitive risk minimization (inverse frequency class weighting & `scale_pos_weight` alignment) to address severe historical class skewness (90.1% Crisis vs. 9.9% Safe observations). ### Pipeline Architecture The project is structured as a sequential, reproducible workflow: - `01_Data Cleaning.ipynb`: Handles raw data ingestion, time-series formatting, primary missing value imputations, and sub-national population capacity estimations (`admin1_pop_est`). - `02_EDA.ipynb`: Conducts exploratory data analysis, mapping data missingness, and calculating early correlation heuristics. - `03_Check Data Pipeline.ipynb`: Validates lag feature engineering (t-1 to t-3), checks for chronological integrity, and ensures zero data leakage. - `04_Model Training.ipynb`: Deploys automated grid searches for Logistic Regress …