Development of an early warning model for stability evaluation of polyphenol-rich fruit juice based on multiple light scattering combined with artificial neural network

• Novel construction of a fruit juice intelligent stability warning model via MLS - ANN integration. • Achieve 7 - day prediction of 90 - day juice stability. • Precisely predicts sedimentation, transmittance, particle and stability index of fruit juice over 90 days. The stability of fruit juice has consistently been an important concern in the food processing industry, which can be time-consuming and costly. Therefore, developing accurate stability early-warning model may serve as a viable solution. Based on Multiple Light Scattering (MLS) technology, this paper collects the stability data as the training set of Triphala fruit juice (TFJ) over a three-month period and finds that the sediment amount reached 0.6 mg/ml, composed of ellagic acid and phlobaphene, with the solution's light transmittance fluctuating range (23%-76%) and the particle size (0.27 μm to 0.29 μm) on day 75. The early warning model comprises a synergistic integration of long short-term memory (LSTM) and backpropagation (BP) neural network models. The model exhibits a mean absolute percentage error (MAPE) of 0.626%, an R 2 of 0.911, and an accuracy of 85.71%. This model is capable of predicting key stability parameters, including sedimentation, transmittance, particle size, particle migration rate, and stability index, within a 90-day period in just 7 days, and thereby provide accurate early-stage stability alerts.