Abstract
Lonicera caerulea L. - well - known in Poland as Kamchatka berry, has been gaining increasing popularity in recent years. The tests carried out on newly established Japanese haskap clones aimed at demonstrating the suitability of the fruits for mechanical harvesting and storage. This study focused on evaluation of mechanical properties and assessment of three distinct machine learning techniques to create predictive models that elucidate the connection between key mechanical attributes of the fruit and storage conditions of L. caerulea. The average force needed to puncture the fruits skin and flesh of L. caerulea var. emphyllocalyx varieties is 16.91% higher than that required for the tested L. caerulea var. kamtschatica varieties. L. caerulea var. emphyllocalyx fruits exhibited a significantly higher respiration rate, with C₂H₄ and CO₂ levels during storage being 25.5% and 10.5% higher, respectively, compared to L. caerulea var. kamtschatica varieties. The machine learning algorithms tested yielded accurate models for deformation and energy prediction. The mean absolute percentage error (MAPE) of these models was determined to be between 14.87 and 20.65%. Models with significantly lower accuracy were obtained for force prediction, with the MAPE reaching 28.95% for L. var. kamtschatica fruit and 42.33% for L. emphyllocalyx fruit. The cultivation and improvement of Lonicera caerulea L. varieties is of great importance for the advancement of mechanized harvesting methods and the development of improved storage technologies for this species. The creation of machine learning methods will facilitate the development of predictive models that can serve as a predictive tool for the relationship between selected mechanical properties