Short-series Prediction with BEMA Approach: application to short rainfall series

• Published in: Revista IEEE América Latina Vol. 14; no. 8; pp. 3892 - 3899
• Main Authors: Rivero, C.R., Pucheta, J.A., Baumgartner, J.S., Laboret, S.O., Sauchelli, V.H., Patino, H.D.
• Format: Journal Article
• Language: English; Spanish
• Published: Los Alamitos IEEE 01.08.2016; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Bayes methods; Bayesian analysis; Bayesian enhanced modified approach; Belief networks; Benchmark testing; complete and incomplete datasets; Data mining; Entropy; Forecasting; IEEE transactions; Mathematical model; Missing data; permutation entropy; Permutations; Rainfall; short rainfall series; Time series
• ISSN: 1548-0992; 1548-0992
• DOI: 10.1109/TLA.2016.7786377

This paper contributes with short time series prediction for complete and incomplete datasets based on a new framework by means of Bayesian enhanced modified approach (BEMA) combining permutation entropy. The focus of the proposed filter with particularly interest in incomplete datasets or missing data is by changing the structure of the predictor filter according to data model selected, in which the Bayesian approach can be combined with entropic information of the series. The simplest method adopted to imputing the missing data on the dataset is by linear average smoothing, then computational results are evaluated on high roughness time series selected from benchmark series, in which they are compared with artificial neural networks (ANN) nonlinear filters such as Bayesian Enhanced approach (BEA) and Bayesian Approach (BA) proposed in recent work, in order to show a better performance of BEMA filter. These results support the applicability of permutation entropy in analyzing the dynamic behavior of chaotic time series for short series predictions.