Determining the Best Extreme Learning Machine Architecture for Indonesian Inflation Forecasting

Abstract

Inflation is an indicator of a country's economic condition. Countries that have low inflation are countries with strong economies. High inflation is an indication that the economy is in trouble. Therefore, inflation forecasting is important to prevent economic problems in the country. Several studies show that machine learning algorithms are more efficient than econometric techniques. One of the machine learning algorithms for forecasting is the Extreme Learning Machine (ELM), namely an Artificial Neural Network (ANN) which has the advantages of the smallest training error, smallest weight norm, best performance and can be run very quickly. In ANN, determining the network architecture will determine forecasting accuracy. Therefore, the aim of this research is to determine the Extreme Learning machine architecture for inflation forecasting in Indonesia. The methods used are data collection, data preprocessing, data exploration, dividing data into training data and testing data, designing the ELM architecture, estimating ELM model weights for each architecture, calculating MSE and determining the best architecture for inflation forecasting. The result of this research is that the best model for Indonesian inflation data is the ELM model with input inflation lag 1, 2 and 5, Eid Al-Fitr indicator variables, and government policy indicator variables, the activation function is sigmoid and the number of neurons in the hidden layer is 10.