Food-Insecurity-Prediction

Project Overview

This project develops machine learning models to predict IPC (Integrated Food Security Phase Classification) severity levels (1–5) using multi-dimensional data, including socio-economic indicators, conflict data, agricultural metrics, and climate factors.

Technical Stack

Data Processing

• pandas >= 1.5.0
• numpy >= 1.24.0

Machine Learning

• scikit-learn >= 1.3.0
• tensorflow >= 2.13.0 #

Visualization

• matplotlib >= 3.6.0
• seaborn >= 0.12.0

Dataset

Overview

This project utilizes a unique, multi-source real-world dataset on food insecurity, combining several authoritative international datasets to build a comprehensive predictive model of food insecurity severity.

Data Sources

Source	Provider	Variables
IPC (Integrated Food Security Phase Classification)	Humanitarian Data Exchange (HDX)	Target variable (IPC Phases 1–5)
World Bank Development Indicators	World Bank	Socio-economic variables (population, GDP, poverty rates)
FAOSTAT	Food and Agriculture Organization of the UN	Agricultural production (maize, rice, wheat, sorghum, millet yields & production), climate indicators (temperature change, variability)
UCDP	Uppsala Conflict Data Program	Conflict-related fatalities (battle-related deaths, one-sided violence against civilians)

Dataset Characteristics

• Temporal Coverage: 20+ years (2003–2024)
• Geographic Scope: Multiple countries, focusing on food-insecure regions with IPC assessments
• Dataset Size: 927–1,200 country-year observations
• Features: Dozens of explanatory variables per observation
• Adequacy: Dataset size is sufficient for robust machine learning experimentation

Data Integration

The dataset represents a careful integration of multiple authoritative sources, providing a comprehensive view of factors contributing to food insecurity across different dimensions:

• Food Security Assessment** (IPC classifications)
• Economic Indicators** (World Bank metrics)
• Agricultural Production** (FAO statistics)
• Conflict Data** (UCDP records)

Data Dictionary: Final IPC-All Clean Dataset

This document describes the variables included in the cleaned dataset used for food insecurity prediction.

Column	Description
iso3	Country 3-letter ISO code
country	Country name
year	Year of observation
phase_type	IPC analysis type: current, proj1 (first projection), proj2 (second projection)
ipc_phase	IPC Phase classification (1=Minimal, 2=Stressed, 3=Crisis, 4=Emergency, 5=Famine)
population_total	Total population (World Bank, SP.POP.TOTL)
gdp_current_usd	Gross Domestic Product in current US dollars (World Bank, NY.GDP.MKTP.CD)
poverty_rate	Poverty headcount ratio at $2.15/day (% of population, World Bank, SI.POV.DDAY)
brd_deaths	Battle-related deaths (UCDP Battle-related deaths dataset)
one_sided_deaths	One-sided violence deaths (UCDP One-sided violence dataset)
Crop: Maize (corn) - Production	Maize (corn) total production (tonnes)
Crop: Maize (corn) - Yield	Maize (corn) yield (hg/ha)
Crop: Rice - Production	Rice total production (tonnes)
Crop: Rice - Yield	Rice yield (hg/ha)
Crop: Wheat - Production	Wheat total production (tonnes)
Crop: Wheat - Yield	Wheat yield (hg/ha)
Crop: Millet - Production	Millet total production (tonnes)
Crop: Millet - Yield	Millet yield (hg/ha)
Crop: Sorghum - Production	Sorghum total production (tonnes)
Crop: Sorghum - Yield	Sorghum yield (hg/ha)
Climate: Standard Deviation	Climate indicator: Standard Deviation
Climate: Temperature change	Climate indicator: Temperature change

Target Variable

ipc_phase: Severity classification (1-5)

• 1: Minimal food insecurity
• 2: Stressed
• 3: Crisis (acute food insecurity)
• 4: Emergency (humanitarian emergency)
• 5: Famine (humanitarian catastrophe)

Key Identifiers

• iso3: Country ISO3 code
• country: Country name
• year: Assessment year
• phase_type: Assessment type/period

Team Members & Section Assignments

Section	Team Members	Key Responsibilities
1. Data Description & Sources	Elizabeth, Mable, Immaculate	Dataset overview, feature documentation, data dictionary
2. Data Cleaning & Missing Values	Conrad, Denis	Data preprocessing, missingness handling, clean dataset creation
3. Exploratory Data Analysis	Mable, Nicholas	Univariate/multivariate analysis, visualization, patterns
4. Classical ML Models	Charles, Nicholas	Traditional ML implementation, baseline models, evaluation
5. Neural Network Models	Conrad, Immaculate	Deep learning implementation, advanced modeling
6. Findings & Conclusions	Charles, Denis, Conrad	Results synthesis, presentation, final documentation

Project Workflow

Section 1: Data Description & Sources (Elizabeth, Mable, Immaculate)

Objectives:

-Brief the target: ipc_phase (1–5) classification -Identify the key identifiers: iso3, country, year, phase_type -Describe feature categories: socio-economic, conflict, agriculture, climate -Create comprehensive data dictionary with feature definitions -Document data sources and collection methodology

Data Cleaning & Missing Values (Conrad, Denis)

Cleaning Steps:

-Inspect missingness - Calculate percentage missing per column -Drop high-missingness columns - Remove columns with >70% missing values

Imputation strategy:

• Numeric features: Median imputation

• Conflict deaths: Fill NaN with 0 (assume no reported fatalities)

• Optional time-series imputation: Per-country forward-fill for GDP/population

• Save cleaned dataset - Export df_model.csv for modeling & EDA

Exploratory Data Analysis (Mable, Nicholas)

Univariate Analysis:

-Target distribution: Bar chart of ipc_phase frequencies -Feature distributions: Histograms of GDP, poverty, population, climate indicators -Summary statistics: Central tendency and spread measures -Outlier detection: Identify extreme values per feature

Multivariate Analysis:

-Correlation matrix: Numeric feature relationships heatmap -Target relationships: Scatter plots of GDP vs ipc_phase, Poverty vs ipc_phase -Conflict analysis: Conflict deaths vs ipc_phase visualization -Feature comparisons: Boxplots of ipc_phase vs key numeric features -Geographic patterns: Country-level analysis and mapping

Classical ML Models (Charles, Nicholas)

Model Development:

-Data preparation: Stratified train/test split by ipc_phase

Baseline models:

-Logistic Regression (multiclass) -Random Forest Classifier -SVM

-Feature engineering: Create relevant derived features

Model evaluation:

-Primary metrics: Accuracy, Macro-F1 score -Secondary metrics: Precision, Recall per class -Confusion matrices: Detailed classification analysis

Neural Network Models (Conrad, Immaculate)

Deep Learning Implementation:

• Architecture design: Simple MLP (Multi-Layer Perceptron) classifier
• Output layer: Softmax activation for multiclass classification
• Model training: Optimization and regularization techniques
• Performance evaluation: Compare against classical models using Macro-F1
• Model interpretation: Feature importance and prediction analysis

Findings & Conclusions (Charles, Denis, Conrad)

Analysis Questions:

-Feature importance: Which features mattered most for predictions? -Model comparison: Which model performed best and why? -Risk assessment: IPC coverage limitations, data missingness, potential bias -Practical implications: Real-world application feasibility -Future improvements: Next steps and recommendations

How to run the project

• Click on this Google Colab link: [https://colab.research.google.com/drive/1I0MBIFvNsYu-vMGSQAsUVIOn2ZNOGyqI]
• Sign in to your Google account if prompted
• If the dataset isn't automatically loaded, upload the CSV file:
• Click on the folder icon in the left sidebar
• Click "Upload" and select your CSV dataset file
• Click "Runtime" → "Run all" to execute all cells Or run cells individually by clicking the play button on each cell

Authored By Immaculate,Conrad,Mable,Nicholas,Denis,Elizabeth,Charles