AN APPLICATION OF GREY MODEL IN MODELLING SOCIO-ECONOMIC VARIABLES WITH LIMITED DATA IN NORTH KALIMANTAN

Abstract

New administrative regions often face a severe data scarcity precludes the use of conventional econometric models and data-intensive machine learning techniques. This study evaluates the performance of Grey System Theory, specifically the classical GM(1,1) and the extended model EXGM(1,1) in modeling &forecasting socio-economics variables under limited data conditions. Utilizing official time-series data from 2013–2021, models were developed using an 8-year training set and validated against a 1-year testing set. Performance was measured using the Mean Absolute Percentage Error (MAPE). Mathematical findings reveal that both models achieve highly accurate performance with MAPE under 5% for variables with near-monotonic trends, such as Gross Regional Domestic Product (GRDP) and unemployment rate. While EXGM(1,1) demonstrated superior mathematical fit during the training phase due to optimized background values. The classical GM(1,1) proved more resilient during the testing phase, particularly for volatile indicators like poverty rates. Specifically, EXGM(1,1) exhibited a risk of overfitting to grey noise in non-monotonic datasets, leading to higher forecasting errors compared to the GM(1,1). It concludes that while EXGM(1,1) offers superior flexibility for tracking turning points in stable decaying trends, the classical GM(1,1) remains the more reliable tool for general policy interpolating due to its predictive stability and resistance to stochastic fluctuations.