An Empirical Study of Time Series Forecasting Using Boosting Technique with Correlation Coefficient

One of the most important fields of researches and applications is time series forecasting. The task to find a model that can fit the data is not easy, because the most of the problems the series are complex and noisy. Recently, ensemble of machines had been used to get accurate predictions. The mean idea is to combine predictions from different forecast methods in only one predictor and in this way to improve the accuracy. This paper explores genetic programming (GP) and Boosting technique to obtain an ensemble of predictors and proposes a new approach to the Boosting algorithm where the correlation coefficients are used to update the weights and the final hypothesis instead of the loss function used traditionally by the boosting algorithm. To validate this method, experiments were accomplished using real and artificial series generated by Monte Carlo simulation. The results obtained by using this new methodology was compared with the results obtained from GP, GPBoost and the traditional statistical methodology (ARMA).