Ground beef shelf-life prediction

Food waste is a growing global issue that impacts consumers, producers, and environmental sustainability. Food insecurity is also a critical issue, with an estimated 900 million people affected by severe food insecurity, globally, in 2022. The production of high-quality food products, including beef products, in the US may alleviate this; however, many of those meat products that are produced tend to be discarded before they can be consumed. In 2024, an estimated one-fifth of the food that was produced globally was wasted. A primary reason for this elevated food waste at the consumer and retail level is the use of overly conservative sell-by and use-by dates on packages. These dates offer a variation of product descriptions with assumptions that they relate to food safety, though the products are generally still good days after. Consumers believe that these dates are definitive markers of food safety of the product and will discard it according to those conservative dates. Therefore, it is important that these dates accurately express the product's shelf-life, so it is not discarded prematurely. Meat spoilage is driven by numerous factors, but primarily by microbial activity. The bacteria that populate the surface of the meat break down the tissue and produce the chemical compounds that cause off-odors and flavors of spoiled meat, as well as discoloration, slime, and gas production. The goal of this research project was to monitor the changes in microbial community structures over a 14-day shelf-life and apply machine learning to generate a practical model of spoilage for predicting shelf-life. Throughout the 14-day sampling, data was collected through colorimeter testing, lipid oxidation testing, microbial aerobic plate counts, and meat compositional analysis. The overall microbial communities were assessed using amplicon sequencing methods and DNA extraction was performed using the Illumina miSeq platform. A statistical assessment of the microbial taxonomies was used to see how they change over time. Following this, a Random Forest regression machine learning algorithm was constructed to predict the day the product is considered microbiologically and sensory spoiled. The quality analysis results show that the product was microbiologically spoiled at day 6 of the experimental period, with spoilage being defined as a microbial aerobic plate count (APC) of 7 logCFU/g. Further evidence of spoilage included the HunterLab colorimeter data that demonstrated decreasing L* values as spoilage progressed, meaning the lightness of the cherry red color was becoming darker. Additionally, the redness values decreased throughout spoilage as oxidation occurred. The results from lipid oxidation also demonstrate spoilage as the values of mg/kg of malonaldehyde increase. As we continue our analysis, the microbial communities involved became less diverse as certain organisms tend to outcompete the others in the environment. Prior to spoilage, the higher concentration of communities such as Rhodobacteraceae and Enterobacterales, and as spoilage progressed, families Pseudomonadaceae and Carnobacteriaceae increased. There was also a decrease in the number of microbial communities present throughout the shelf-life as the organisms outcompeted each other in the environment. This pattern provided data to help with predicting the rate of product spoilage, ultimately aiming to decrease global food waste.