Machine Learning and Taguchi DOE Combined Approach for Modeling Dynamic Ultrasound-Assisted Fresh-Cut Leafy Green Sanitation

Chlorine-based fresh produce sanitation is a dynamic process, and sanitation efficiency is limited due to chlorine degradation. Here, ultrasound was coupled with a benchtop sanitation system to enhance chlorine sanitizer efficiency in fresh-cut leafy green sanitation. Taguchi design of experiments (DOE) and machine learning (ML) were combined to model the relationship between sanitation condition parameters and sanitation outcomes. Multiple ML algorithms were fitted, tuned, and compared for performance using 127 experimental trials (training-to-validation ratio = 3:1). Gaussian process regression (GPR) models showed the best performance in predicting sanitation outcomes of chemical oxygen demand (COD, R-2 = 0.73), remaining Escherichia coli O157:H7 on the leaf surface ("Surface Microbe", R-2 = 0.88), and E. coli O157:H7 concentration in sanitation water ("Water Microbe", R-2 = 1.00). Cut size and agitation speed were identified as the most critical input parameters. An initial free chlorine concentration over 20 mg/L was recommended to minimize the E. coli O157:H7 concentration in sanitation water. This work showcases the combined approach of ML and DOE in optimizing fresh-cut produce sanitation. Moreover, it provides a solution for overcoming the difficulties of modeling multiple controllable and uncontrollable factors with reduced experimental runs.