Use of Boosting Algorithms in Household-Level Poverty Measurement - A Machine Learning Approach to Predict and Classify Household Wealth Quintiles in the Philippines

Aug 5, 2026

This study assesses the effectiveness of machine learning models for predicting household poverty levels in the Philippines using five boosting algorithms: AdaBoost, CatBoost, GBM, LightGBM, and XGBoost. CatBoost delivers the strongest performance, reaching about 91% accuracy and the highest precision, recall, and F1-score, while XGBoost and GBM follow with slightly lower results. The work also evaluates computational efficiency through training time, testing speed, and model size, showing that CatBoost has longer training but high testing efficiency. Findings indicate machine learning can support poverty prediction and more targeted policy interventions, with future work recommending broader datasets to improve predictive accuracy and policy usefulness.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

AccessFree to download

Introduction

Poverty measurement and policy relevance

Types of poverty metrics: monetary vs multidimensional

Limitations of econometric approaches

Machine learning for refined poverty measurement

Highlights Use of Boosting Algorithms in Household-Level Poverty Measurement: A Machine Learning Approach to Predict and Classify Household Wealth Quintiles in the Philippines Erika Lynet V. Salvador • CatBoost achieved the highest accuracy (90.93%) and overall performance metrics, followed by XGBoost, GBM, and LightGBM. AdaBoost had the lowest performance. • AUC-ROC scores indicated that CatBoost, GBM, LightGBM, and XGBoost excelled in distinguishing between poverty classes, while AdaBoost lagged behind. • Computational efficiency varied, with AdaBoost having the shortest training time but longest testing time. CatBoost had the longest training time but was highly efficient during testing. GBM, LightGBM, and XGBoost balanced well between training and testing times. arXiv :2407 . 13061v1 [ cs .CY] 28 May 2024 Use of Boosting Algorithms in Household-Level Poverty Measurement: A Machine Learning Approach to Predict and Classify Household Wealth Quintiles in the Philippines Ms. Erika Lynet V. Salvadora,b,∗,1 (Researcher) a Department of Mathematics and Statistics, Amherst College, College St., Amherst, 01002, Massachusetts, United States of America b Science, Technology, Engineering, and Mathematics Strand, Senior High School Department, De La Salle University Integrated School Manila, Taft Avenue, Malate, Manila, 2401, National Capital Region, Philippines ARTICLE INFO Keywords: machine learning classification household wealth index Philippines AB STRACT This study assessed the effectiveness of machine learning models in predicting poverty levels in the Philippines using five boosting algorithms: Adaptive Boosting (AdaBoost), Cat Boosting (CatBoost), Gradient Boosting Machine (GBM), Light Gradient Boosting Machine (LightGBM), and Extreme Gradient Boosting (XGBoost) . CatBoost emerged asthe superior model and achieved the highest scores across accuracy, precision, recall, and F1-score at 91%, while XGBoost and GBM followed closely with 89% and 88% respectively. Additionally, the research examined the computational efficiency of these models to analyze the balance between training time, testing speed, and model size—factors crucial for realworld applications. Despite its longer training duration, CatBoost demonstrated high testing efficiency. These results indicate that machine learning can aid in poverty prediction and in the development of targeted policy interventions. Future studies should focus on incorporating a wider variety of data to enhance the predictive accuracy and policy utility of these models. 1. Introduction As of 2024, over 700 million people globally live in extreme poverty and survive on less than $2.15 (Php 125) per day [1] . To address this, governments worldwide are intensifying efforts to achieve Sustainable Development Goal 1 (SDG) which targets the eradication of poverty in all its forms by 2030 . However, recent research suggests that the lingering effects of the COVID-19 pandemic may endure in various countries until 2030 [2] . This presents a significant challenge to the goal of reducing global poverty—a target already at risk prior to the crisis. Hence, the need for significant political intervention is now more pressing than ever [1] . In order for policymakers to formulate targeted interventions and allocate resources efficiently, accurately determining poverty levels is paramount [3] . Data empowers governments and organizations to devise strategies that genuinely uplift individuals and communities from poverty. Without accurate data, policy initiatives risk falling short in addressing the underlying causes of poverty or reaching those most in need [4] . Poverty, however, is defined diversely. Broadly, poverty measurement approaches fall into two categories: monetary and non-monetary [5] . The monetary approach, as the name suggests, defines poverty based on income or expenditure. For instance, the established poverty methodology in the Philippines employs pre-tax i

FAQs

Which boosting algorithms were compared for predicting poverty levels?

The study compares AdaBoost, CatBoost, GBM, LightGBM, and XGBoost for household-level poverty measurement in the Philippines.

What model performed best and how were its results summarized?

CatBoost achieved the highest accuracy (about 90.93%) and top overall performance metrics, with the highest precision, recall, and F1-score near 91%. XGBoost and GBM ranked next, while AdaBoost had the lowest performance.

How did the algorithms differ in computational efficiency?

Training and testing efficiency varied across models: AdaBoost showed the shortest training time but the longest testing time, while CatBoost had the longest training time but strong testing efficiency. GBM, LightGBM, and XGBoost balanced training and testing time more evenly.

Aug 5, 2026

Upload by Clementine