Machine learning-guided optimization of probiotic chickpea milk fermentation enhances antioxidant capacity, volatile aroma profile, and sensory quality

Abstract

This study developed and optimized a probiotic chickpea milk fermentation process using Lactiplantibacillus plantarum PC4, integrating response surface methodology with artificial neural network genetic algorithm (ANN-NSGA-II) modeling. Fermentation parameters (time, temperature, inoculum level) were systematically evaluated for their effects on antioxidant activity (DPPH, FRAP, TPC), pH, and probiotic viability. Although response surface models showed strong fitting accuracy (R² ' 0.96), ANN demonstrated superior prediction stability and captured nonlinear behavior across all responses (Test R² ' 0.97). Multi-objective optimization using NSGA-II maximized antioxidant capacity and viable cell counts while maintaining pH near 4.5. The optimal compromise solution (7.40 h, 35.95 °C, 2.22 %) achieved high antioxidant values and 8.81 log CFU/mL viable probiotics, with validation experiments demonstrated reasonable agreement between the predicted and experimental results. Fermentation enhanced β-glucosidase activity, lactic acid accumulation, and the formation of alcohols and esters, while markedly reducing aldehyde-driven beany odor. Sensory evaluation confirmed stronger fermented aroma, lower beany notes, and improved overall flavor. These findings establish ANN-NSGA-II as an effective and precise framework for developing functional plant-based fermented beverages.