Eni Sumarminingsih, Rahma Fitriani, Dwi Ayu Lusia, Aqsa Yudhistira Redi, Natasha Aulia
Inflation is a disease-like economic condition that affects nearly every country worldwide. High inflation rates indicate that a country’s economic condition is less than ideal. Therefore, controlling inflation is crucial for maintaining a healthy and robust national economy. To control inflation it is necessary to forecast accurate inflation. The modeling for forecasting inflation is quite complex if it is done using existing statistical models because there are many predictor variables that influence inflation. A better approach to use is a machine learning approach. Long Short Term Memory (LSTM) is a machine learning approach capable of producing the best model for inflation forecasting. Therefore, in this study, LSTM will be used as a method to model inflation forecasting. The added value of this study lies in using a comprehensive variable or indicator that determines inflation, so that before modeling using the LSTM, data exploration will be carried out first to determine the indicators that affect inflation. To determine the input, economic theory is used which is supported by statistical analysis, namely correlation analysis and partial autocorrelation function (PACF) statistics. In this research, five input designs were tried. The most optimal LSTM model in predicting inflation in Indonesia is a model that utilizes input variables in the form of Indonesian inflation at lags 2 and 4, the CNY, EUR, and USD exchange rates, interest rates, Eid al-Fitr indicators, government policy indicators, and world oil prices. © 2025 Eni Sumarminingsih et al.
Department of Statistics, Universitas Brawijaya, Jalan Veteran, Malang, Indonesia
