NON-DESTRUCTIVE ESTIMATION OF TENDER COCONUT WATER VOLUME VIA IMPACT-INDUCED USING BIO-INSPIRED OPTIMIZED REGRESSION MODELS

• A non-destructive acoustic framework is developed for accurate estimation of tender coconut water volume. • Vibrational attenuation exhibits a strong, monotonic relationship with internal water content, independent of impact orientation. • Bio-inspired optimized ensemble models substantially outperform classical regression and neural approaches in prediction accuracy and stability. • The method demonstrates high robustness to knocking position, enabling reliable measurements without strict positional control. • The proposed approach supports low-cost, portable implementation for real-time coconut quality assessment in field conditions. • Bio-inspired optimized models significantly improved prediction accuracy, with GWO-GBR achieving RMSE = 4.38 and R 2 ≈ 0.996. A low-cost, non-destructive estimation of coconut water volume remains a critical challenge for quality assessment and grading in post-harvest processing. A bio-inspired optimized vibration framework was developed for non-destructive estimation of tender coconut water volume. The approach utilizes vibration attenuation features obtained from an Arduino-based sensing system and models their relationship with internal water content using machine learning techniques. The dataset was partitioned into training (80%) and testing (20%) sets using stratified sampling, and model robustness was further validated using k-fold cross-validation. A strong correlation between attenuation and water volume was observed (Pearson r ≈ 0.97), confirming the physical relevance of the proposed index. Among the evaluated models, the Grey Wolf Optimizer (GWO)-tuned Gradient Boosting model achieved the best performance (RMSE = 4.38, MAE ≈ 2.19, R 2 ≈ 0.99), followed by ABC and DE tuned Random Forest models (RMSE ≈ 4.64). Bio-inspired optimization significantly enhanced predictive accuracy, with up to ∼60% reduction in RMSE compared to baseline models. The effect of knocking position on prediction accuracy was minimal, and selected optimizers demonstrated stable and rapid convergence. The proposed framework enables accurate, repeatable, and computationally efficient estimation of coconut water volume, demonstrating strong potential for integration into compact, field-deployable quality assessment systems.