Garcia-Lopez Yvan J, Del Carpio Castro Luis A
CENTRUM Católica Graduate Business School (CCGBS), Lima, Peru.
Pontificia Universidad Católica del Perú (PUCP), Lima, Peru.
PLoS One. 2025 Feb 25;20(2):e0318813. doi: 10.1371/journal.pone.0318813. eCollection 2025.
This study addresses the challenges of measuring regional competitiveness using traditional methods, due to the inherent complexity and non-linearity of its determinants'. The development of new Machine Learning (ML) models allows the creation of predictive models capable of handling this type of data, providing actionable insights. The objective of the study was to develop and test the use of non-linear Machine Learning models to measure the regional competitiveness in Peru, at the sub-national level. The research uses the ODD (Overview, Design Concepts, and Details) protocol to ensure a transparent and replicable methodology. The impact of ML on the Peruvian Regional Competitiveness Index (IRCI) is examined across 25 regions from 2016 to 2023, focusing on five key pillars: economy, government, infrastructure, businesses, and people. A suitability index (IoI) was developed to assess how well the pillar components align with ML. Data provided by CENTRUM PUCP was subjected to exploratory data analysis (EDA) to address variability among pillar scores and their effects on competitiveness. Six nonlinear machine learning models (Gradient Boosting, Random Forest, XGBoost, AdaBoost, Neural Networks, and Decision Trees) were applied, and the machine learning models with the highest predictive accuracy were Gradient Boosting and Random Forest. Performance metrics include MSE values of 1.1399 and 1.3469, RMSE values of 1.0677 and 1.1606, and R2 values of 0.9768 and 0.9729, respectively. These results demonstrate the effectiveness of machine learning in analyzing the complexity of regional competitiveness data, identifying influential variables, and reducing score distortions. The findings provide a data-driven framework for policymakers to improve regional competitiveness, which promotes academic knowledge and practical applications for sustainable development.
由于区域竞争力决定因素具有内在的复杂性和非线性，本研究探讨了使用传统方法衡量区域竞争力所面临的挑战。新的机器学习（ML）模型的发展使得能够创建能够处理这类数据的预测模型，并提供可操作的见解。本研究的目的是开发和测试使用非线性机器学习模型来衡量秘鲁次国家层面的区域竞争力。该研究采用ODD（概述、设计概念和细节）协议，以确保方法的透明性和可重复性。从2016年到2023年，对秘鲁25个地区的25个地区进行了考察，重点关注五个关键支柱：经济、政府、基础设施、企业和人民。开发了一个适用性指数（IoI）来评估支柱组成部分与机器学习的匹配程度。对CENTRUM PUCP提供的数据进行了探索性数据分析（EDA），以解决支柱得分之间的差异及其对竞争力的影响。应用了六种非线性机器学习模型（梯度提升、随机森林、XGBoost、AdaBoost、神经网络和决策树），预测准确率最高的机器学习模型是梯度提升和随机森林。性能指标包括MSE值分别为1.1399和1.3469，RMSE值分别为1.0677和1.1606，R2值分别为0.9768和0.9729。这些结果证明了机器学习在分析区域竞争力数据的复杂性、识别有影响的变量和减少得分扭曲方面的有效性。研究结果为政策制定者提供了一个数据驱动的框架，以提高区域竞争力，促进学术知识和可持续发展的实际应用。
