A Machine Learning Study About the Vulnerability Level of Poverty in Perú

Silva Marchan H.A.; Peña Cáceres O.J.M.; Ricalde Moran D.M.; Samaniego-Cobo T.; Perez-Espinoza C.M. · Communications in Computer and Information Science (2022) · DOI: 10.1007/978-3-031-19961-5_1

Resumen

In recent years, people have been requiring new livelihoods that allow them to have enough economical resources for the development of their daily activities, considering the problematic that COVID-19 has brought to their lives. The objective of this research was to analyze machine learning algorithms such as Decision Tree, Random Forest, Naive Bayes, Logistic Regression and Vector Support Machine, in order to identify the risk level to fall into poverty for a person in Perú, basing the analysis on the National Household Survey (NHS) that the National Institute of Statistics and Informatics (NISI) provided on 2020. The methodology was presented in four stages, organization and structuring of the database, analysis and identification of the variables, application of the learning algorithms and evaluation of the performance of the aforementioned algorithms. Python programming language and the STATA software allowed the exploration of 91,315 registers and 33 variables. Results showed that the Decision Tree algorithm has an accuracy of 98%, while other algorithms are below the indicated range, so dynamism is expected in the application of this algorithm for socioeconomic areas that can be materialized through a permanent evaluation and analysis platform that helps to focus strategies and proposals for the benefit of the population with economic limitations.

Instituciones: Universidad Nacional de Piura; Universidad Nacional de Tumbes