Improving kinetic model fitting for total titratable acidity in bananas using genetic algorithms

This research aimed to automate the fitting of kinetic models using genetic algorithms (GAs) to optimize the estimation of kinetic parameters—the reaction rate constant (k) and the initial value of total titratable acidity (TTA, C₀)—and enhance predictive accuracy. Experimental data were collected from bananas (Musa paradisiaca L.) at ripening stage 2 on Von Loesecke scale, measuring TTA at specific intervals, and three kinetic models—zero-order, first-order, and second-order—were fitted. For optimization, Python-based GA was used to minimize the mean squared error (MSE) and evaluate performance based on R2, AIC, and BIC. Results indicated that while the zero-order model provided the best statistical fit (R2 = 0.918, lowest AIC and BIC), validation on unseen data favored the second-order model, which exhibited superior predictive accuracy (test R2 = 0.9349, lowest test MSE). The integration of GAs enhanced the robustness of kinetic fitting by effectively navigating parameter space and avoiding local minima, outperforming traditional optimization methods, demonstrating their ability to refine parameter estimation and improve model generalization. Future research should explore hybrid models that combine mechanistic kinetic equations with machine learning techniques (e.g., neural networks, or support vector regression), to better capture the biochemical processes underlying acidity changes in bananas. DOI: https://doi.org/10.54167/tch.v19iEspecial.1847