Identifying graft incompatible rootstocks for sweet cherry through machine learning algorithms

Abstract

Graft incompatibility is a key factor in the development of dwarf and semi dwarf rootstocks for sweet cherry (Prunus avium L.) to improve yield, fruit quality, precocity, and labor efficiency. This study evaluated the graft incompatibility of eight genotypes three sweet cherry, three sour cherry, and two mahaleb collected from Northern Anatolia, a native region for cherries. These genotypes, along with standard rootstocks Gisela 6 and SL 64, were grafted with ‘0900 Ziraat’ and ‘Lambert’ cultivars. Graft incompatibility was assessed using a multidisciplinary approach combining classical morphological and anatomical evaluations with advanced data driven analyses. Parameters such as graft bud growth rate (40.26–86.21%), shoot length (41.01–91.28 cm), and rootstock/scion diameter ratio (0.41–0.92) were measured 12 months after grafting. Principal Component Analysis, Random Forest modeling with SHAP values, and Bayesian ranking were applied to identify key traits and rank genotype performance. The integrated analysis successfully distinguished compatible rootstock candidates, identifying five genotypes with high compatibility potential. These findings demonstrate that combining traditional phenotypic evaluation methods with machine learning-based approaches offers a robust and comprehensive framework for addressing graft incompatibility, and contributes valuable insights for future breeding programs and rootstock selection strategies in sweet cherry.