Commodity price forecasting via machine learning models

The prediction of values in a time series is the object of study in several fields of knowledge. In the agricultural commodities futures market, this type of information can be used to minimize risks to investments and contribute to the increase in the volume of negotiations of various commodities. As the prices of these assets are influenced by many external varia-bles, predictions are usually made through analysis fundamentalist or technical and this work is carried out by specialists in the field. That restricts the access of individuals who could invest, but does not do so because they do not have this knowledge that is necessary for the survival of this business on the Stock Exchanges. This article approaches computational methods, which involve the algorithms: k-nearest neighbor; random forest; Artificial neural networks; support vector machine; extreme gradient boosting; and two meta-models, ensemble by average and stac-king, applied to historical data of the following commodities: sugar; live cattle; coffee; ethanol; corn; and soybeans with the objective of predicting prices in horizons of one and ten steps ahead using the regression technique. Forecast errors are measured using statistics from the MAPE error metric, which demonstrates that the support vector machine is the algorithm with the best accuracy for the analyzed series. The results indicate that the predictions of intelligent models have high performance in the short term. In this sense, speculators and hedgers can benefit from using the proposed technique, as a support for decision making.