Forecasting hog prices with multisource heterogeneous data: A secondary data-decomposition based approach

Hog prices are a topic of concern for stakeholders such as hog farmers, pork-related food companies, and investors. For them, accurately predicting hog prices and understanding the importance of influencing factors are necessary for many related decisions. In order to achieve the above goals, this study attempts to use an explainable artificial intelligence (XAI) approach to forecast hog prices with multisource heterogeneous data and secondary data decomposition. Firstly, economic indexes and commodity prices, search engine data (SED), and online news related to hog prices are collected. Secondly, kernel principal component analysis is used for dimension reduction, sentiment analysis is conducted for online news, and two decomposition methods are adopted to decompose historical hog price into various components. Then, the extracted kernel principal components, market sentiment index, and decomposed components are used as feature inputs in optimized LightGBM models. Finally, forecasting performance of different models is compared, and model interpretation is analyzed. Using China’s weekly average hog futures price data, empirical analysis shows that both multisource heterogeneous data and secondary data decomposition contribute to improving predictive accuracy, and the proposed model outperforms other models. The interpretability analysis further addressed that secondary decomposition components and market sentiment index are important features in predicting hog price. The proposed XAI approach is the contribution in AI, and the adjustment of hog production and slaughter based on price forecasts and interpretability analysis is the application in engineering.