Análise comparativa da eficiência de três modelos de previsão de séries temporais do software Weka

Abstract

This work aims to compare three time series forecasting models from the WEKA software: the Linear Regression (LR), the Support Vector Machine (SMOreg) and the Artificial Neural Network (MLP). The database, expanded by interpolation with python, was obtained from the Instituto Nacional de Meteorologia (INMET). This database presents a historical series of wind speed, at the meteorological station of Palmeira dos Índios, in the period between 2008 and 2020. Forecast models, based on neural networks MLP, Support Vector Machine and Linear Regression, were implemented in WEKA, a free software widely used in academic research. Results obtained from the three models were compared using the metrics RMSE (Root Mean Squared Error), MAPE (Mean Absolute Percent Error) and MAE (Mean Absolute Error). It was verified, for a short term horizon (14 instances), that the Neural Network Multilayer Perceptron performed better.