From Bytes to Bites - Using Country Specific Machine Learning Models to Predict Famine

Aug 3, 2026

The research investigates how machine learning can predict and support decisions during famine and hunger crises, focusing on household nutrition as measured by food consumption scores. Using natural, economic, and conflict-related variables, the study trains three regression models—Linear Regression, XGBoost, and RandomForestRegressor. RandomForestRegressor achieves the lowest average prediction error (10.6%), with accuracy varying by country. Economic indicators are consistently the most influential predictors, while no single feature generalizes across all regions. The findings support country-specific modeling and improved data collection to strengthen global hunger forecasting.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages17

Reading Time43 min

AccessFree to download

I. Introduction

Historical and traditional forecasting frameworks

Machine learning for hunger crisis decision support

Project approach and objectives

II. Background

Prior machine learning work and transferable features

XGBoost-based food security transition forecasting

From Bytes to Bites: Using Country Specific Machine Learning Models to Predict Famine Salloni Kapoor1 , Simeon Sayer2 1 Issaquah High School, Issaquah, WA, 98027 , United States 2 Faculty of Arts and Sciences, Harvard University, Cambridge, MA, 02138, United States [kapoorsalloni5@gmail.com](kapoorsalloni5@gmail.com); [ssayer@fas.harvard.edu](ssayer@fas.harvard.edu) Hunger crises are critical global issues affecting millions, particularly in low-income and developing countries. This research investigates how machine learning can be utilized to predict and inform decisions regarding famine and hunger crises. By leveraging a diverse set of variables—natural, economic, and conflict-related—three machine learning models (Linear Regression, XGBoost, and RandomForestRegressor) were employed to predict food consumption scores, a key indicator of household nutrition. The RandomForestRegressor emerged as the most accurate model, with an average prediction error of 10.6%, though accuracy varied significantly across countries, ranging from 2% to over 30%. Notably, economic indicators were consistently the most significant predictors ofaverage household nutrition, while no single feature dominated across all regions, underscoring the necessity for comprehensive data collection and tailored, country-specific models. These findings highlight the potential of machine learning,particularly Random Forests, to enhance famineprediction, suggesting that continued research and improved data gathering are essential for more effective global hunger forecasting. I. Introduction Famine has plagued human societies throughout history, and despite advances in technology and data analysis, predicting and preventing hunger crises remains a significant global challenge. Throughout the ages, people have turned to traditions and cultural references, such as Groundhog Day in the U.S. and Canada, to make sense of unpredictable conditions and forecast future food supply. More theoretical approaches have also been used as justification, such as the theories proposed by Thomas Malthus [1], suggesting that famine was inevitable as population growth would eventually surpass Earth’s food supply. These ideas provided early frameworks for understanding food scarcity, acting as stepping stones for modern systems like FEWSNET (Famine Early Warning Systems Network) [2] which offer real-time data on food security. Despite their analytical approach, such systems still struggle with accurately predicting future conditions. In recent years, machine learning has emerged as a promising tool for predicting and informing decisions regarding hunger crises. Machine learning models, particularly those using supervised learning and regression techniques, have the potential to enhance famine predictions by analyzing vast amounts of data and identifying patterns that might be missed by traditional methods. However, many of these approaches fail to fully integrate the available data for the region they are analyzing, creating weaker models. Additionally, the results of these models give little guidance into region specific priorities for improving future data collection, which is critical for refining and enhancing predictive accuracy. This project builds on existing approaches for predicting future food insecurity by employing a regression model that leverages numerical data, including satellite imagery, food specific economic factors, and conflict-related statistics. Unlike previous papers that often focus solely on a specific region, or utilize a limited set of input variables, this approach integrates abroad array of data sources, but methodically selects the richest subset of that data to train a region specific model. This research proves that there is no universal formula for famine, and that important variables in one region are useless in another. By pairing these optimal regional inputs with the corresponding food consumption score [3]—a key indicator of

FAQs

What is the main goal of the project described in the document?

To use machine learning to predict famine and hunger crises by modeling household nutrition outcomes, represented by food consumption scores.

Which machine learning models are used, and which performs best?

The study uses Linear Regression, XGBoost, and RandomForestRegressor; RandomForestRegressor is the most accurate with an average prediction error of 10.6%.

How do predictors vary across countries, and what role do economic indicators play?

Accuracy differs widely by country (about 2% to over 30%), and while economic indicators are consistently the most significant predictors, no single feature dominates across all regions.

Aug 3, 2026

Upload by Theodora