Application of LASSO Regression for the Identification of Underdeveloped Regions in Central Sulawesi

Abstract

This study aims to identify the main factors influencing regional underdevelopment in Central Sulawesi through Human Development Index (HDI) modeling and to develop a robust predictive model. To address the challenges of multicollinearity and the limited number of observations (13 districts/cities with 10 variables), this study employs LASSO (Least Absolute Shrinkage and Selection Operator) regression, which is capable of simultaneously shrinking coefficients and selecting variables. The data used are sourced from the 2019 publication of the Central Statistics Agency (BPS). The analysis was conducted using descriptive statistics, Ordinary Least Squares (OLS) modeling, VIF tests, and LASSO regression with cross-validation (leave-one-out cross-validation). The results indicate that very high multicollinearity (VIF > 10 for most variables) renders the OLS model unstable. Conversely, LASSO regression yielded better performance with superior RMSE (1.282), MAE (1.075), and R² (0.918) values compared to OLS (RMSE 21.67; MAE 9.85; R² 0.78). Thus, LASSO is more suitable for limited data with high multicollinearity. The selected significant variables include the percentage of the poor population, the open unemployment rate, shopping facilities, the presence of hospitals, the population density ratio, and the number of elementary and secondary schools.