Accurate hybrid prediction model for poverty line, number, and percentage of impoverished individuals

Abstract

Poverty remains a major social issue in many developing countries, including Indonesia, as seen in the Central Java region. Over the last five years, the number of impoverished people in Central Java has shown fluctuations, with data from the Central Statistics Agency indicating figures of 3,897.20 thousand (2018), 3,743.23 thousand (2019), 3,980.90 thousand (2020), 4,109.7 thousand (2021), and 3,831.44 thousand (2022). Analyzing these trends is crucial for future poverty reduction efforts. This study aims to develop a web-based predictive system capable of forecasting the poverty line, as well as the number and percentage of poor residents in Central Java. The research utilizes a hybrid forecasting model that integrates the Holt-Winters triple exponential smoothing (HWTES) method with fuzzy time series (FTS), alongside algorithmic approaches such as rate of change (RoC) and frequency-based segmentation. The model's accuracy, evaluated using the average absolute percentage error (MAPE), shows low error rates: 0.9% for the number of impoverished people, 1.6% for the percentage, and 0.7% for the poverty threshold. Compared to the standard HWTES model, this hybrid model demonstrates greater precision. As a result, it can serve as an effective tool to support strategic planning and enhance poverty alleviation programs in Central Java.