Prediction for hog prices based on similar sub-series search and support vector regression

Abstract Predicting hog price is important for making decisions for administration sections and pig-breeding enterprises. Hog prices follow a time series that is non-stationary, non-linear and has a pseudo-period, and that changes as a result of potential growth, cyclical fluctuation and errors. Considering the different characteristics of the trend component and the cyclical component in prediction, in this paper, we propose a hog price prediction method to address the problem of pseudo-cycle caused by the varying cycle length. We begin by separating the cyclical component and the trend component of the hog price series. We then predict the cyclical component of hog price series using a most similar sub-series search method, and predict the trend component using support vector regression. Finally, we combine the predicted series. Our main contributions are proposing a method that predicts the cyclical and trend components of hog prices separately, and designing a most similar sub-series search method to predict the cyclical component. In experiments on real datasets, our method has minor errors and exhibits superior performance compared with existing methods. It is suitable for predicting the price series of hog and other agricultural products with similar characteristics.