Predição de séries de indicadores agropecuários por método de clustering não supervisionado

Abstract

This work proposes the development of a prediction system for the agricultural indicators series based on the use of a clustering method, Clustering, subjecting the model to a unsupervised training. For this, they are used as serial input data, corresponding to the actual closing prices, spot prices of agricultural products, the Stock Exchange of São Paulo considered as first-line values. The result obtained by the model is compared to the prediction generated by two reference methods (Naive Network and MLP). Analyzes of the resulting predictions of the model and the other predictors, which will be shown to be superior to the Naive model method and similar to the MLP network for context prediction of one step ahead will be performed. At the end of this work, proposals for future work are suggested