<![CDATA[Central Sulawesi experiences fluctuating poverty levels despite various government programs, making accurate forecasting essential for data-driven policies. This study aims to model and forecast the number of poor people in Central Sulawesi by comparing linear, non-linear quadratic, and non-linear cubic trend regression models based on data from Badan Pusat Statistik (BPS) 2002–2025. Initially, these three models were evaluated using the Ordinary Least Squares (OLS) method, where the quadratic and cubic models yielded relatively similar values of 80.65% and 80.76%, respectively. However, rigorous diagnostic tests revealed that the residual independence assumption under OLS was severely violated across all models due to time-series dependencies. To resolve this issue, the Generalized Least Squares (GLS) method was employed to account for underlying autocorrelation structures. The linear GLS model ( = 552.6687 - 8.3413t) significantly outperformed both the quadratic and cubic GLS variants; it was the only model that fully satisfied the normality and autocorrelation-free (white noise) assumptions while maintaining a high predictive accuracy with a MAPE of 5.484%. Projections using this optimal linear model indicate a stable, continuous downward trend, predicting the poor population to decrease to 344.17 thousand in 2026 (t = 25) and eventually reach its lowest historical point of 227.41 thousand by 2040 (t = 39).]]>
Copyrights © 2026
