Interval price prediction of livestock product based on fuzzy mathematics and improved LSTM.

Livestock product prices serve as a barometer and bellwether for the agricultural market. However, traditional point prediction techniques focus mainly on tracking or fitting, resulting in limited information and challenges in evaluating the uncertainty of future prices. A comprehensive livestock price prediction model with joint point and interval prediction capabilities is proposed, with fuzzy mathematics and long short-term memory. Three main steps are taken: (1) data composition and reconstruction, to extract a set of relatively stationary subsequence components by complementary ensemble empirical mode decomposition (CEEMD) from original signal, and divide these components into three groups according to fuzzy entropy (FE) value. (2) characteristics categorization, determining the lower bound, mean, and upper bound of the rebuilt data via fuzzy information granulation (FIG) to better characterize the price fluctuation range. (3) price prediction, including point and interval predictions with attention mechanism long short-term memory (AM-LSTM). An empirical study was conducted on the weekly price data of pork, beef, and mutton in China from 2009 to 2023, incorporating discussions on different embedding dimensions, prediction step, fuzzy granulation window sizes, decomposition techniques, and prediction algorithms. The results indicate that the proposed interval prediction model can not only achieve high accuracy in point prediction, but also better capture price change intervals.