A novel ISSA-GRNN approach for efficient camphor extraction from Artemisia argyi

Efficient extraction of bioactive compounds from plant matrices is essential for pharmaceutical, functional food, and agricultural applications. However, process optimization is often challenged by nonlinear interactions among multiple parameters and limited experimental data. This work presents an optimization framework that combines an improved sparrow search algorithm (ISSA) with a generalized regression neural network (GRNN) to model camphor extraction from Artemisia argyi leaves. The ISSA incorporates three modifications: hybrid chaotic-opposition population initialization, a dual-adaptive discoverer update rule, and a stagnation-triggered hybrid mutation operator. The GRNN leverages its small-sample tolerance to establish a nonlinear relationship between five extraction parameters and camphor yield using 50 experimental runs. Under 5-fold cross-validation, ISSA-GRNN outperformed PSO, GA, and SSA-optimized GRNN models, achieving an R² of 0.931, RMSE of 0.0140, and MAPE of 4.65%. Experimental validation under optimized conditions (58 °C, 0.7% cellulase, 3.8 h hydrolysis, pH 6.0, 32 min ultrasonication) produced a camphor yield of 0.2368 mg/g, an 11.5% increase over RSM, with an RSD of 1.05% (3.74% for RSM). The findings of this study contribute to a better understanding of small-sample nonlinear process modeling and provide a practical framework for bioactive compound extraction under data-scarce conditions, readily extendable to other plant matrices and data-limited regression problems.