Advanced machine learning streamlines market-timing decisions

Deciding when to bring a pig to market has never been an easy task.

To maximize profits, farmers must evaluate changes in animal weight, pork prices, feed costs and pen space, while maintaining an inventory of market-ready pigs to meet long-term contractual obligations with meat processors. With so many changing variables, farmers face what researchers call the “curse of dimensionality”: having too much data to best solve a problem analytically. It is therefore practically impossible to obtain optimal policies to solve such problems.

However, research co-authored by a professor at UC’s Riverside School of Business leverages artificial intelligence (AI) to break this curse. Using price and inventory data from a large Illinois hog operation, researchers developed a machine learning model to determine when, to whom and how many pigs to sell to maximize profits in the long term. The model recovered about 22% of the profits that farmers typically lose with traditional decision-making.

The article is published in the SSRN Electronic Journal.

“Traditional farming methods tend to focus on immediate profits, neglecting how today’s choices affect future income. This myopia can mean missed opportunities and lower overall profits in the long term,” he said. said Danko Turcic, associate professor of operations and supply chain at UCR. direction and co-author of the article.

Turcic explained that farmers face critical decisions once pigs reach the “finishing stage,” an age of about six months when pigs reach a market-ready weight of about 200 pounds. This is a time when farmers must decide how many pigs to sell and how many they will continue to feed to meet their future contractual obligations or perhaps extract higher profits from larger pigs on the open market if they anticipate a rise meat prices.

To help farmers make such decisions, Turcic and his co-authors started with an AI model commonly used in computer games and adapted it to the world of pig farming. This involved adding realistic constraints, such as limits on the number of pigs that could be sold at a time due to contracts with meatpackers, and ensuring that the AI ​​would not try to sell more pigs than what the farmer actually had.

Above all, they ensured that the AI ​​decision-making process was transparent and understandable. This “interpretability” is crucial in fields like agriculture and medicine, where users must trust AI recommendations.

By understanding how AI worked, researchers discovered why it was much better than current practices for selling pork. AI could identify the best times to sell, such as when market prices were high enough to offset penalties for breaking a contract. He also strategically held back some pigs, anticipating either higher prices later or periods when there would be fewer pigs available to sell.

Although the study focuses on pig farming, its implications extend to all sectors involving perishable products and dynamic inventory management. Turcic suggests that similar systems could optimize decisions in agriculture, retail, and even the launch timing of consumer products such as smartphones.

“Our work highlights the potential of AI to not only automate but also augment human decision-making with powerful tools that were previously unimaginable,” Turcic said.

The study, titled “An Empirical Analytical Approach to Managing the Finishing Phase of Pig Farms,” is available online and has been accepted for publication in the Operations Management Journal. His co-authors are Panos Kouvelis of the Olin Business School at Washington University in St. Louis and Ye Liu of the Martin J. Whitman School of Management, Syracuse University, Syracuse, NY.