Comparative analysis of mathematical models and machine learning approaches for predicting dragon fruit seed mass

This study evaluates the performance of traditional empirical models and machine learning approaches, specifically artificial neural networks (ANN), for predicting the mass of dragon fruit ( Hylocereus spp. ) seeds. Physical parameters of seeds were measured using a digital vernier caliper (P-VC) and a computer vision system with image processing (P-CV-IP). Mass prediction was conducted through mathematical models (MM-P-VC and MM-P-CV-IP) and ANN-based models (ANN-P-CV-IP). Results demonstrated that ANN models outperformed traditional mathematical approaches, achieving the highest coefficient of determination (R 2 ) and the lowest root mean square error (RMSE), particularly in volume-based predictions. Among mathematical models, those based on physical dimensions provided the most reliable outcomes. Both P-VC and P-CV-IP measurements proved suitable for mass estimation; however, image-based quadratic models offered practical accuracy for routine applications. Digital image analysis further enabled rapid, non-destructive assessment of seed size and mass, while also capturing additional morphological attributes. Overall, the integration of image processing with ANN provides a precise, efficient, and resource-saving approach for dragon fruit seed mass modelling, offering potential for broader applications in agricultural product characterization. • Comparison was made between mathematical models and artificial neural networks (ANN) for predicting dragon fruit seed mass efficiency and accuracy. • ANN-based models, particularly volume-based, exhibit superior predictive performance. • Mathematical mass models for dragon fruit seed mass prediction, with physical dimensions models exhibited promising results. • Physical parameters of dragon fruit seeds proved effective in predicting weight characteristics, utilizing both vernier calliper and computer vision measurements. • Integration of image processing and ANN in mass modelling offers a non-invasive, efficient, and accurate method for predicting seed mass.