Pronóstico de la inflación colombiana : una aproximación desde los modelos machine learning

Abstract

The objective of this paper is to forecast monthly Colombian inflation based on its macroeconomic determinants. 7 machine learning models are used: linear regression, SMV, Decision Trees, MLP, KNN, SVR and LSTM, and 1 conventional ARIMA model. The models with the best prognosis were the ARIMA and the LSTM. Although, the prediction of the LSTM can be improved by making an optimal architecture of the data since it manages to capture the drastic changes of the variables, it could even be improved if the behavior of each of the divisions that make up the basic basket is included.