ABSTRACT:

Rapidly evolving businesses generate huge amounts of it slow-stamped data sequences and cause a requirement for each univariate and multivariate time series forecasting. For such data, ancient predictive models primarily based on autoregression are usually not sufficient to capture complicated nonlinear relationships between multidimensional options and so the time series outputs. In order to take advantage of these relationships for improved time series forecasting whereas additionally better addressing a wider choice of prediction situations, a forecasting system requires a flexible and generic design to accommodate and tune various individual predictors also combination ways in which. In reply to the current challenge, an design for combined, multilevel time series prediction is proposed, that is appropriate for many completely completely different universal regressors and combination methods. The key strength of this design is its ability to build a diversified ensemble of individual predictors that form an input to a multilevel choice and fusion methodology before the final optimized output is obtained. Excellent generalization ability is achieved because of the highly boosted complementarity of individual models additional enforced through cross-validation-linked training on exclusive data subsets and ensemble output postprocessing. In a sample configuration with basic neural network predictors and a mean combiner, the proposed system has been evaluated in numerous eventualities and showed a transparent prediction performance gain.