Analisis Tingkat Kemiskinan di Indonesia Menggunakan Model Vanilla Long Short-Term Memory dan Stacked Long Short-Term Memory

Abstract

Indonesia, as the fourth most populous country in the world and a developing nation, faces significant challenges in addressing widespread poverty. Poverty is a condition where individuals or groups have limited access to adequate economic resources, quality food, healthcare services, and education. Despite government efforts to implement programs aimed at reducing poverty levels in Indonesia, these programs have often been ineffective and poorly targeted. The objective of this research is to compare the performance of two Long Short-Term Memory (LSTM) models, Vanilla LSTM and Stacked LSTM, in analyzing poverty levels in Indonesia. The data used for this study is from the year 2021 and encompasses 514 cities across Indonesia. This data includes variables such as the percentage of the impoverished population, regional gross domestic product, life expectancy, average years of schooling, and per capita expenditure, all of which are relevant to Indonesia's economic and social conditions.The research employs Vanilla LSTM and Stacked LSTM models. Evaluation is conducted using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Normalized Root Mean Squared Error (NRMSE), and Mean Absolute Error (MAE) as the main metrics to measure the accuracy of the model predictions. The results indicate that Vanilla LSTM consistently outperforms Stacked LSTM, achieving an MSE of 0.0109, RMSE of 0.1046, NRMSE of 0.1334, and MAE of 0.0795. In contrast, Stacked LSTM shows an MSE of 0.0119, RMSE of 0.1091, NRMSE of 0.1391, and MAE of 0.0833. These findings suggest that Vanilla LSTM has lower and more stable prediction errors and is more accurate in estimating poverty levels. Vanilla LSTM is therefore a better choice for analyzing and reducing poverty levels in Indonesia. This model can serve as an effective tool for policymakers to design more efficient and targeted strategies to reduce poverty rates.