Study Finds Explainable Machine Learning Predicts Grape Shelf Life in Storage

A study published in Food Chemistry says explainable machine learning can help predict how long table grapes will keep in storage and how their anthocyanin levels will change over time, offering a more detailed way to track fruit quality as it declines.

The work focused on Red Globe grape berries and examined one of the main problems in postharvest handling: fruit decay does not happen in a simple or uniform way. Changes in texture, acidity, color and internal atmosphere can develop together and at different speeds, which makes shelf-life hard to estimate with standard methods. The researchers said a predictive system that can also explain which variables matter most could support both preharvest decisions in the field and postharvest storage management.

To build those predictions, the researchers treated grapes with five commercial salts and compared them with a water control. The salts were calcium chloride, sodium chloride, potassium acetate, sodium acetate and sodium bicarbonate. The berries were then stored at 0°C and 20°C to generate different patterns of quality loss and anthocyanin accumulation during storage.

Anthocyanins are the pigments that give red and purple grapes much of their color. They are also closely tied to visible ripening and market appeal. In the study, anthocyanin accumulation was analyzed alongside shelf-life because both are important to how grapes change after harvest and during storage.

The research team fed several kinds of information into four machine learning algorithms. The data included chemical composition, fruit quality measurements and the surrounding gas atmosphere, specifically O2 and CO2. According to the study, the goal was not only to produce accurate predictions but also to understand why the models reached those predictions.

Among the four algorithms tested, random forest regression performed best. The study said those models showed potential to predict both shelf-life and anthocyanin accumulation using either chemical parameters or gas atmosphere composition. That matters because growers, packers and storage operators often need tools that can work with the kind of measurements already collected during handling and storage.

The study also used SHAP values, a method designed to show how much each input feature contributes to a model’s prediction. This is the “explainable” part of the approach. Rather than treating machine learning as a black box, SHAP analysis can point to the variables that carry the most weight in forecasting changes in fruit quality.

In this case, the SHAP results highlighted titratable acidity and anthocyanin as especially important in predicting the shelf-life of stored grapes. That finding suggests acidity and pigment development are not just general quality markers but central signals in how the model estimates whether stored grapes are approaching the end of their useful commercial life.

The researchers said the models can help distinguish which factors influence quality decay and anthocyanin accumulation, which could guide more targeted preharvest and postharvest strategies. In practice, that could mean adjusting treatments, storage conditions or monitoring routines based on the variables that the model identifies as most influential.

Although the study was carried out on table grapes, the work could also draw attention from the beverage sector. Color development linked to anthocyanins is a key issue in wine grapes, and storage or maturation decisions often depend on the same kind of quality indicators examined here. The researchers did not study wine production, but the monitoring and prediction approach could potentially be adapted to grape lots used for winemaking or to other fruit-based beverage supply chains where ripeness, color and postharvest stability affect processing decisions.

The study presents the work as part of a broader move toward data-driven management in horticulture. By combining storage data with machine learning models that can be interpreted, the researchers said the method opens a path to fruit storage systems that do more than record decline after it happens, and instead help anticipate it while fruit is still moving through the supply chain.